Advances in Control, Signal Processing and Energy Systems: Select Proceedings of CSPES 2018 [1st ed. 2020] 978-981-32-9345-8, 978-981-32-9346-5

This book comprises select proceedings of the National Conference on Control, Signal Processing, Energy and Power System

523 84 15MB

English Pages XXII, 229 [240] Year 2020

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Advances in Control, Signal Processing and Energy Systems: Select Proceedings of CSPES 2018 [1st ed. 2020]
 978-981-32-9345-8, 978-981-32-9346-5

Table of contents :
Front Matter ....Pages i-xxii
Front Matter ....Pages 1-1
Anti-windup Control of Nonlinear Cascade Systems with Particle Swarm Optimization Parameter Tuning (Fernando Serrano, Josep M. Rossell)....Pages 3-16
Pollutant Profile Estimation Using Unscented Kalman Filter (S. Metia, S. D. Oduro, A. P. Sinha)....Pages 17-28
Determination of Model Order of an Electrochemical System: A Case Study with Electronic Tongue (Sanjeev Kumar, Arunangshu Ghosh)....Pages 29-38
Front Matter ....Pages 39-39
Problem Diagnostic Method for IEC61850 MMS Communication Network (Anjali Gautam, S. Ashok)....Pages 41-54
IntelliNet: An Intelligence Delivery Network (Dipnarayan Das, Sumit Gupta)....Pages 55-66
A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System (Sumit Gupta, Puja Halder)....Pages 67-77
Object Detection in Clustered Scene Using Point Feature Matching for Non-repeating Texture Pattern (Soumen Santra, Partha Mukherjee, Prosenjit Sardar, Surajit Mandal, Arpan Deyasi)....Pages 79-96
Human Behavior Recognition: An l1 – ls KSVD-Based Dictionary Learning and Collaborative Representation-Based Classification (Pubali De, Amitava Chatterjee, Anjan Rakshit)....Pages 97-105
Detection and Classification of Breast Cancer in Mammographic Images Using Efficient Image Segmentation Technique (Pramit Brata Chanda, Subir Kumar Sarkar)....Pages 107-117
Front Matter ....Pages 119-119
Visualization and Improvement of Voltage Stability Region Using P-Q Curve (Srijan Seal, Debjani Bhattacharya)....Pages 121-134
Analysis of Temperature at Substrate and Sink Area of 5 W COB-Type LEDs, with and Without Driver (Debashis Raul)....Pages 135-145
Performance Study and Stability Analysis of an LED Driver (Piyali Ganguly, Vishwanath Gupta, Parthasarathi Satvaya)....Pages 147-158
Instrumentation for Wireless Condition Monitoring of Induction Machine (Soumyak Chandra, S. Saruk Mohammad, Rajarshi Gupta)....Pages 159-167
Solar PV Battery Charger Using MPPT-Based Controller (Shreya Das, Avishek Munsi, Piyali Pal, Dipak Kumar Mandal, Sumana Chowdhuri)....Pages 169-182
Comparative Study on Simulation of Daylighting Under CIE Standard Skies for Different Seasons (Abhijit Gupta, Sutapa Mukherjee)....Pages 183-197
Application of Modified Harmony Search and Differential Evolution Optimization Techniques in Economic Load Dispatch (Tanmoy Mulo, Prasid Syam, Amalendu Bikash Choudhury)....Pages 199-213
Design of a Multilevel Inverter Using SPWM Technique (Arka Ray, Shuvadeep Datta, Amitava Biswas, Jitendra Nath Bera)....Pages 215-229

Citation preview

Lecture Notes in Electrical Engineering 591

Tapan Kumar Basu Swapan Kumar Goswami Nandita Sanyal Editors

Advances in Control, Signal Processing and Energy Systems Select Proceedings of CSPES 2018

Lecture Notes in Electrical Engineering Volume 591

Series Editors Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli Federico II, Naples, Italy Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán, Mexico Bijaya Ketan Panigrahi, Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore Rüdiger Dillmann, Humanoids and Intelligent Systems Lab, Karlsruhe Institute for Technology, Karlsruhe, Baden-Württemberg, Germany Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China Gianluigi Ferrari, Università di Parma, Parma, Italy Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid, Madrid, Spain Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität München, Munich, Germany Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA, USA Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt Torsten Kroeger, Stanford University, Stanford, CA, USA Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA Ferran Martin, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra, Barcelona, Spain Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA Sebastian Möller, Quality and Usability Lab, TU Berlin, Berlin, Germany Subhas Mukhopadhyay, School of Engineering & Advanced Technology, Massey University, Palmerston North, Manawatu-Wanganui, New Zealand Cun-Zheng Ning, Electrical Engineering, Arizona State University, Tempe, AZ, USA Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Gan Woon Seng, School of Electrical & Electronic Engineering, Nanyang Technological University, Singapore, Singapore Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Baden-Württemberg, Germany Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China Junjie James Zhang, Charlotte, NC, USA

The book series Lecture Notes in Electrical Engineering (LNEE) publishes the latest developments in Electrical Engineering - quickly, informally and in high quality. While original research reported in proceedings and monographs has traditionally formed the core of LNEE, we also encourage authors to submit books devoted to supporting student education and professional training in the various fields and applications areas of electrical engineering. The series cover classical and emerging topics concerning:

• • • • • • • • • • • •

Communication Engineering, Information Theory and Networks Electronics Engineering and Microelectronics Signal, Image and Speech Processing Wireless and Mobile Communication Circuits and Systems Energy Systems, Power Electronics and Electrical Machines Electro-optical Engineering Instrumentation Engineering Avionics Engineering Control Systems Internet-of-Things and Cybersecurity Biomedical Devices, MEMS and NEMS

For general information about this book series, comments or suggestions, please contact leontina. [email protected]. To submit a proposal or request further information, please contact the Publishing Editor in your country: China Jasmine Dou, Associate Editor ([email protected]) India Swati Meherishi, Executive Editor ([email protected]) Aninda Bose, Senior Editor ([email protected]) Japan Takeyuki Yonezawa, Editorial Director ([email protected]) South Korea Smith (Ahram) Chae, Editor ([email protected]) Southeast Asia Ramesh Nath Premnath, Editor ([email protected]) USA, Canada: Michael Luby, Senior Editor ([email protected]) All other Countries: Leontina Di Cecco, Senior Editor ([email protected]) Christoph Baumann, Executive Editor ([email protected]) ** Indexing: The books of this series are submitted to ISI Proceedings, EI-Compendex, SCOPUS, MetaPress, Web of Science and Springerlink **

More information about this series at http://www.springer.com/series/7818

Tapan Kumar Basu Swapan Kumar Goswami Nandita Sanyal •



Editors

Advances in Control, Signal Processing and Energy Systems Select Proceedings of CSPES 2018

123

Editors Tapan Kumar Basu Indian Institute of Technology Kharagpur Kharagpur, West Bengal, India

Swapan Kumar Goswami Jadavpur University Kolkata, West Bengal, India

Nandita Sanyal B.P. Poddar Institute of Management and Technology Kolkata, West Bengal, India

ISSN 1876-1100 ISSN 1876-1119 (electronic) Lecture Notes in Electrical Engineering ISBN 978-981-32-9345-8 ISBN 978-981-32-9346-5 (eBook) https://doi.org/10.1007/978-981-32-9346-5 © Springer Nature Singapore Pte Ltd. 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Committees

Chief Patrons Shri Arun Poddar, Chairman, B.P. Poddar Foundation for Education and B.P. Poddar Group

Patrons Shri Ayush Poddar, Vice-Chairman, B.P. Poddar Foundation for Education and B.P. Poddar Group Dr. Subir Choudhury, Founder Trustee and Chief Mentor, B.P. Poddar Foundation for Education Prof. (Dr.) Sutapa Mukherjee, Principal, B.P. Poddar Institute of Management & Technology Prof. (Dr.) B. N. Chatterjee, Dean (Academics), B.P. Poddar Institute of Management & Technology

Advisory and Technical Programme Committee Dr. Siddhartha Sen, IIT Kharagpur Dr. N.K. Kishor, IIT Kharagpur Dr. Amitava Chatterjee, Jadavpur University Dr. Abhijit Lahiri, Supreme knowledge Foundation Dr. Samarjit Sengupta, University of Calcutta Dr. Jitendra Nath Bera, University of Calcutta Prof. Sugata Munshi, Jadavpur University Prof. Pranab Kumar Dutta, IIT Kharagpur

v

vi

Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr. Dr.

Committees

Ashoke Sutradhar, IIEST Shibpur Sanjoy Saha, Jadavpur University Debasis Chatterjee, Jadavpur University Kamalika Ghosh, Jadavpur University Kalyan Chatterjee, IIT(ISM) Dhanbad Sovan Dalai, Jadavpur University Diganta Saha, Jadavpur University Biswendu Chatterjee, Jadavpur University Debangshu Dey, Jadavpur University Ranjit Kr. Barai, Jadavpur University Parimal Acharjee, NIT Durgapur Somnath Pan, IIT(ISM) Dhanbad Sanjoy Mondal, IIT(ISM) Dhanbad Saikat Mookherjee, Jadavpur University Arghya Mitra, VNIT Nagpur Kaushik Das Sharma, University of Calcutta Chandan Kr. Chanda, IIEST Shibpur Biswarup Basak, IIEST Shibpur Rajarshi Gupta, University of Calcutta Jayanta Kumar Chanda, Purulia Government Engineering College Madhubanti Mitra, Jadavpur University Subrata Chatterjee, NITTTR Ranjan Kr. Behera, IITP

Organizing Committee

General Chair Prof. Dr. Tapan Kumar Basu, IIT Kharagpur (Retired)

Organizing Chair Dr. Nandita Sanyal, Head, Department of EE

Committees

vii

Technical Programme Chairs Dr. Krishnendu Chakraborty, Principal, Government College of Engineering and Ceramic Technology Dr. Sudipta Chakraborty, Associate Professor, B.P. Poddar Institute of Management & Technology

Finance Committee Dr. Subhasish Pradhan, Registrar, BPPIMT Mr. Amlan Roy Choudhury, BPPIMT Ms. Chandrani Das, BPPIMT Mr. Aritra Ghosh, BPPIMT

Publication Committee Dr. Ivy Majumdar, BPPIMT Dr. Sutapa Mukherjee, BPPIMT Dr. Indrakanta Maitra, BPPIMT Mr. Argha Kamal Pal, BPPIMT

Industry Relations, Registration and Publicity Ms. Ms. Ms. Ms.

Anushree Roy, BPPIMT Susmita Dey, BPPIMT Madhumita Kundu (Mondal), BPPIMT Sujata Saha, BPPIMT

Website Committee Mr. Subhadip Chandra, BPPIMT Mr. Subhasish Das, BPPIMT

Preface

It was a great pleasure with which we released the proceedings of National Conference on Control, Signal Processing & Energy Systems (CSPES 2018) organized by the Department of Electrical Engineering, B.P. Poddar Institute of Management and Technology, Kolkata, India, on 16–18 November 2018. This conference was technically supported by IEEE CSS IMS Joint Chapter Kolkata, WEBREDA, Department of Power and N. E. S West Bengal, and the Institution of Engineering and Technology (IET), Kolkata Network. This conference was a modest effort to assemble academicians, researchers, engineers and technocrats under a tutelage to facilitate the exchange of ideas, thoughts, and research outcomes which would inspire students for higher studies and research and also to find the gaps between existing curriculum and industry requirements. We like to thank all the authors for contributing their manuscripts and all the reviewers whose effort and hard work contributed towards the quality of submissions. The editors wholeheartedly like to acknowledge the constant encouragement and support from the institute management. Our sincere thanks extend to Dr. Akash Chakraborty, Associate Editor, Applied Science and Engineering, Springer. We also like to thank our sponsors for extending their financial support to hold the conference. We are indebted to Dr. Debanshu Dey and his team from Jadavpur University and Dr. Kaushik Das Sharma from the Department of Applied Physics, University of Calcutta, for constant technical support. Last but not least, we are sincerely thankful to all the faculty members, staffs and students for their tireless effort with which the publication of CSPES 2018 Proceedings came true. Kharagpur, India Kolkata, India Kolkata, India

Tapan Kumar Basu Swapan Kumar Goswami Nandita Sanyal

ix

About the Institute

In 1999, B.P. Poddar Institute of Management and Technology (BPPIMT) was set up as a tribute to late B.P. Poddar, a visionary philanthropist, educationist and founding father of the group. Supported by the B.P. Poddar Foundation for Education, a trust dedicated to enriching the quality of technical education in the country, the institute is affiliated to the Maulana Abul Kalam Azad University of Technology (MAKAUT), West Bengal, and approved by the All India Council for Technical Education (AICTE). B.P. Poddar Institute of Management and Technology aims for a better society by bettering the education system. Its ambition resets on its unique learning culture that encourages collaboration and communication and the dedication of its experienced faculty picked from diverse fields. The courses are offered in the disciplines of computer science and engineering, electronics and communication engineering, electrical engineering and information technology. The institute blends a dynamic and progressive approach towards outcome-based education teaching–learning with a vision to emerge as a progressive and premier institute for engineering and technology education with ethical values for creative engineering solutions commensurate with global changes. The mission of the institute is to offer quality education through modern, accessible, comprehensive and research-oriented teaching–learning process; create opportunities for students and faculty members in acquiring knowledge through research and development; provide an effective interface with the industry by strengthening industry–institute interaction and develop entrepreneurial skills; and meet ever-changing needs for the nation through rational evolution towards sustainable and environment-friendly technologies. The institute management also always encourages to create a platform for professional development activities in the institute in association with professional engineering societies/chapters and to help students to organize and participate in invited lectures, workshops, seminars and other technical events to improve

xi

xii

About the Institute

technical skills. The institute also tries to bridge the gap between the institute and the industry, thus enhancing the relationship among each other. The aim of the institute is to make an effective contribution to the educational system by identifying the gap between academic curriculum and need of the industry.

Keynote Speakers

Prof. (Dr.) Sutapa Mukherjee, Patron It is my pleasure to inform that Electrical Engineering department of B.P. Poddar Institute of Management and Technology is organizing a National Conference on Control, Signal Processing and Energy System during 16th to 18th November, 2018 which will provide a platform for the researchers and students to share their ideas and to enrich the knowledge through interaction with the experts in their respective domains. As Head of the Institute, I welcome you all to this Conference to make it a grand success. Prof. (Dr.) Sutapa Mukherjee, Patron, CSPES 2018

Prof. Dr. Tapan Kumar Basu, General Chair Respected Chief Guest Mr. Samar Roy and Today’s Guest of Honour & Keynote Speaker in the morning session Prof. Shivaji Chakraborty, Principal Prof. Dr. Sutapa Mukherjee, our Ex-Dean Prof. Biswanath Chatterji my, dear faculty colleagues of BPPIMT, and members of the staff, delegates from different institutions, my dear student friends, ladies and gentlemen, A very Good Morning to all of you! On behalf of the organizing committee of CSPES2018 and on my personal behalf, I extend a Hearty Welcome to all of you to this 3-day National event. The idea of holding a conference is to create an opportunity for young scholars for intellectual discourse and exchange notes in their professional areas and get exposed to newer ideas through brain storming sessions.

xiii

xiv

Keynote Speakers

I am confident that the young researchers will get a platform here to interact with other scholars and at the end will get immensely enriched when they go back. I thank all of you for coming over here to participate in this modest programme; though it is being the third technical conference organized by the Department of Electrical Engineering. I hope you will excuse us for any lapse on our part. Prof. Dr. Tapan Kumar Basu, General Chair, CSPES 2018 Dr. Nandita Sanyal, Organizing Chair It is a great pleasure for me to be a part of organising Committee of the National Conference on Control, Signal Processing and Energy Systems CSPES2018 and to welcome the participants from all over India, to exchange experience and share new ideas for these three days. It is worth mentioning that Joseph Maria Roselle from Spain is also participating in this conference. I verily welcome all participants of CSPES 2018 and wish to thank our Chief Mentor and founder Trustee Dr. Subir Chowdhury, Principal Dr. Sutapa Mukherjee, Former Dean academics Prof. B. N. Chatterjee, and Registrar Dr. Subhasis Pradhan for their patronage and active support. I am also thankful to Prof. Tapan kumar Basu, Prof. Krishnendu Chakraborty and Prof. Sudipta Chakraborty who kindly consented to become General and Program Chairs. I am also thankful to Dr. Ivy Majumdar from Department of ECE and Amlan Roy Chowdhury from Department of CSE. We have the privilege of having Prof. Sivaji Chakravorti Director NIT Calicut as the Guest of Honour and Keynote Speaker in the Conference. I am also thankful to Mr. Samarendra nath Roy former Director BHEL India as Chief Guest. All the Invited Speakers, Session Chairs from renowned Universities have showed their honour to me for giving consent to participate in the conference in spite of their busy schedule. Team work of all the Faculty members from Department of Electrical Engineering make this Dream come true. I am thankful to them. We have got immense support from IET Kolkata network, Globsyn on financial aspects and WEBREDA, CSS IMS Joint Chapter IEEE Kolkata section in technical aspects. This conference proceedings will be published by Springer LNEE series. Finally, this conference is for the Students. If outcome of this conference can inspire them for higher studies, Research and to become successful professional the purpose will be served. “Na chor haryam, Na cha raj haryam Na bhratu bhajyam Na cha bharkariVyaye krute vardhart ev nityaamVidya dhanam sarva dhane pradhanam”. No one can steal it, not authority can snatch, Not divided in brothers, not heavy to carry, As you consume or spend, it increases; as you share, it expands, Knowledge (Vidhya) is the best wealth among all the wealth anyone can have. Hope CSPES 2018 will be a grand Success. Dr. Nandita Sanyal, Organizing Chair, CSPES 2018

Keynote Speakers

xv

Prof. K. Chakrabarty, Program Chair On behalf of the Program Committee, it is my great pleasure to welcome you to the National Conference on Control, Signal processing and Energy system (CSPES 2018) organized by the Electrical Engineering Department of B.P. Poddar institute of management and technology, Kolkata. CSPES 2018 brings together researchers to discuss the latest advances in Control, Signal processing and Energy system. This will also throw light on the direction of research and development in those areas that are very essential for the development of the civilization. The Technical Program of CSPES 2018 consists of tutorials, symposia, keynote addresses, industry sessions and exhibitions. The keynote speakers will highlight the state-of-the art advancements in control, signal processing and other emerging topics in energy systems. Together, all these forums present cutting-edge advances of both the scientific and industrial developments in modern engineering. The main symposia of CSPES 2018 received many paper submissions from the country and abroad, out of which 20 papers have been accepted. All papers have undergone a rigorous peer review process—every symposium paper was reviewed by at least 3 independent experts, with many receiving even more reviews. In addition to the main symposia, CSPES 2018 features tutorial on emerging and important topics in the field, which will be held on the first day of the conference. A large number of proposals for the tutorials were carefully scrutinized with only half of submitted proposals were finally accepted. Most of the technical symposia papers will be presented in lecture style, while some papers will be presented in interactive sessions for in-depth discussions among respective authors and the conference attendees. The quality of papers in lecture-style and interactive sessions is the same. The only criterion to assign a paper to an interactive session is topic homogeneity. Under the current policy of the conference, all papers must be presented by their authors, which will increase the discussions and lead to fruitful technical exchanges. I would like to especially thank the General Chair-Prof. Tapan Kumar Basu, Organising Chair-Prof. Nandita Sanyal, Technical Program Chair-Prof. Sudipta Chakraborty, Track Chairs-Prof. Aparajita Sengupta, Prof. Kumardeb Banerjee, Prof. Swapan Kumar Goswami, Prof. Tapan Kumar Basu, Prof. Gautam Bandyopadhyay and as well as the all members of the advisory and technical program committee and the external reviewers for their dedication. Without their help, this conference would not be possible. I would also like to thank the Keynote Speakers for contributing to this important part of the program. I look forward to welcome you all in Kolkata, the city of Joy. Prof. K. Chakrabarty, Program Chair, CSPES 2018

xvi

Keynote Speakers

Dr. Sudipta Chakraborty, Technical Programme Chair CSPES 2018, National Conference on Control, Signal Processing and Energy Systems is a humble endeavour to amass scientists, academicians, researchers, engineers and technocrats under an aegis to facilitate exchange of novel ideas, thoughts, research outcomes which would provide impetus to stalwarts and amateurs equally to contribute in innovative breakthroughs in science, technology, engineering and mathematics. This forum will not only be a knowledge hub for veterans but it will equally benefit the entire scientific fraternity and inspire students for higher studies and research. Research papers in three broad domains of Electrical Engineering, namely Control Systems, Signal Processing and Energy systems have been accepted. Since the first two areas are extensively interdisciplinary ones, there has been ample scope for researchers involved in allied arenas to contribute significantly. All the papers have undergone blind peer review before acceptance, based on quality, originality, technical content and relevance. All accepted papers presented at the Conference will be included in Proceedings to be published by Springer. This Conference is hosted jointly by the parent Institute, BPPIMT, in collaboration with IET (The Institution of Engineering and Technology). We are honoured to have IEEE Joint CSS-IMS, Kolkata Chapter as our Technical Co-Sponsor. The esteemed association of WEBREDA, Department of Power and NES, Govt. of West Bengal as technical collaborator, has added further value to this technical meet. Besides technical paper presentation in the three tracks mentioned, the Conference will witness invaluable technical deliberations and addresses by distinguished Professors as well as dedicated research scientists. The programme will be graced by Dr. Sivaji Chakravorti, Director, National Institute of Technology, Calicut and Dr. Shrabani Ghosh, DRDO Lab, Balasore as well as Dr. Kuntal Ghosh, Indian Statistical Institute, Kolkata as Key-note speakers. The first day of the Conference has been reserved for pre-Conference tutorial when renowned personnel from WEBREDA, Department of Power and N. E. S. West Bengal will enlighten the students with their practical knowledge garnered through hands on experience. A plant visit has also been organised to provide the students a feel and insight into the intricate work field where they would venture in near future. Hope the success of the Conference will be replicated in the form of quality research and publications. Welcoming everyone to the knowledge fiesta and wishing a great time ahead. With warm regards, Dr. Sudipta Chakraborty, Technical Programme Chair

Reviewers

Prof. Anirban Mukherjee, IIT Kharagpur Prof. Tapan Basu, IIT Kharagpur (retired) and Visiting Faculty, B.P. Poddar Institute of Management and Technology, Kolkata Dr. R. V Sarvadnya, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Prof. Jaya Sil, IIEST Shibpur Dr. S. S Gajre, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Dr. Debaprasad Kastha, IIT Kharagpur Dr. Aurobinda Rout Roy, IIT Kharagpur Dr. Siddhartha Sen, IIT Kharagpur Dr. L. M. Waghmare, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Dr. S. V. Bonde, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Dr. R. R. Manthalkar, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Dr. Balasaheb Patre, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Dr. Kaushik Das Sharma, University of Calcutta, Kolkata Dr. Rakesh Misra, IIT(BHU) Varanasi Dr. Parthasarathi Bera, Kalyani Government Engineering College, Kalyani Prof. Sanjay Talbar, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Prof. Arun Ghosh, IIT Kharagpur Dr. Sudipta Chakraborty, B.P. Poddar Institute of Management and Technology, Kolkata Prof. Raghunath Holambe, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Prof. Amitava Chatterjee, Jadavpur University, Kolkata

xvii

xviii

Reviewers

Dr. A. V. Nandedkar, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Prof. Manesh Kokare, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded Prof. Sanjoy Kumar Saha, Jadavpur University, Kolkata Prof. Binoy Kumar Roy, NIT Silchar

Contents

Control Systems Anti-windup Control of Nonlinear Cascade Systems with Particle Swarm Optimization Parameter Tuning . . . . . . . . . . . . . . . . . . . . . . . . Fernando Serrano and Josep M. Rossell Pollutant Profile Estimation Using Unscented Kalman Filter . . . . . . . . . S. Metia, S. D. Oduro and A. P. Sinha Determination of Model Order of an Electrochemical System: A Case Study with Electronic Tongue . . . . . . . . . . . . . . . . . . . . . . . . . . Sanjeev Kumar and Arunangshu Ghosh

3 17

29

Signal Processing Problem Diagnostic Method for IEC61850 MMS Communication Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anjali Gautam and S. Ashok

41

IntelliNet: An Intelligence Delivery Network . . . . . . . . . . . . . . . . . . . . . . Dipnarayan Das and Sumit Gupta

55

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sumit Gupta and Puja Halder

67

Object Detection in Clustered Scene Using Point Feature Matching for Non-repeating Texture Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Soumen Santra, Partha Mukherjee, Prosenjit Sardar, Surajit Mandal and Arpan Deyasi Human Behavior Recognition: An l1 – ls KSVD-Based Dictionary Learning and Collaborative Representation-Based Classification . . . . . . Pubali De, Amitava Chatterjee and Anjan Rakshit

79

97

xix

xx

Contents

Detection and Classification of Breast Cancer in Mammographic Images Using Efficient Image Segmentation Technique . . . . . . . . . . . . . 107 Pramit Brata Chanda and Subir Kumar Sarkar Energy Systems Visualization and Improvement of Voltage Stability Region Using P-Q Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 Srijan Seal and Debjani Bhattacharya Analysis of Temperature at Substrate and Sink Area of 5 W COB-Type LEDs, with and Without Driver . . . . . . . . . . . . . . . . 135 Debashis Raul Performance Study and Stability Analysis of an LED Driver . . . . . . . . 147 Piyali Ganguly, Vishwanath Gupta and Parthasarathi Satvaya Instrumentation for Wireless Condition Monitoring of Induction Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Soumyak Chandra, S. Saruk Mohammad and Rajarshi Gupta Solar PV Battery Charger Using MPPT-Based Controller . . . . . . . . . . . 169 Shreya Das, Avishek Munsi, Piyali Pal, Dipak Kumar Mandal and Sumana Chowdhuri Comparative Study on Simulation of Daylighting Under CIE Standard Skies for Different Seasons . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Abhijit Gupta and Sutapa Mukherjee Application of Modified Harmony Search and Differential Evolution Optimization Techniques in Economic Load Dispatch . . . . . . . . . . . . . . 199 Tanmoy Mulo, Prasid Syam and Amalendu Bikash Choudhury Design of a Multilevel Inverter Using SPWM Technique . . . . . . . . . . . . 215 Arka Ray, Shuvadeep Datta, Amitava Biswas and Jitendra Nath Bera

About the Editors

Prof. Tapan Kumar Basu obtained his B.Tech (Hons.) in Electrical Engineering and M.Tech in Power System Engineering in 1968 and 1970 respectively from IIT Kharagpur. Subsequently he joined IIT Delhi as a research Scholar and obtained his Ph.D. in Power System Stability. He joined NIT, Kurukshetra (Formerly known as Regional Engineering College) in Nov. 1973 as a lecturer and later as an Astt. Prof. in the Electrical Engg. Deptt. In July 1976 he joined IIT Bombay as an Astt. Prof. in the Electrical Engg. Deptt. and then moved to IIT Kharagpur in April 1980. He became a Professor in July 1985 and retired from IIT in August 2009 to join the Aliah University, Kolkata as the Dean in Sept. 2009. In October 2010 he joined the Institute of Technology and Marine Engg (ITME) near Diamond Harbour as the Director. After completing his term as the Director, he left ITME and joined Academy of Technology (AOT), Adisaptagram, Dt. Hooghly in Feb. 2013 where he served as a senior Professor in Electrical Engineering till July 2017. Prof. Basu served on the Board of Directors of West Bengal State Electricity Transmission Company Limited (WBSETCL) during 2008–2014. Currently he is an Adjunct Professor at AOT and B.P. PODDAR Institute of Management & Technology (BPPIMT), Kolkata. He has been appointed as an Advisor to the Speech and Image Processing Group of CDAC (Centre for Development of Advance Computing), Kolkata. Prof. Basu guided 13 Ph.D. scholars and several M.tech students in areas of Power system Stability and Forecasting, Signal and Image Processing and Speech Processing and taught a large number of subjects to undergraduate and postgraduate classes during his long teaching career. He has developed two video courses on Networks, Signals and Systems and Digital Signal Processing under NPTEL National Project for Technology Enhanced Learning. He has published more than 150 papers in many national and international journals and conferences. He has obtained many awards He is a Life Fellow of the Institution of Engineers (I), System Society of India and Indian Society for Theoretical and Applied Mathematics.

xxi

xxii

About the Editors

Swapan Kumar Goswami is a Professor in Electrical Engineering, Jadavpur University, Kolkata. He has published more than one hundred research papers. Since 2013 his papers were cited 1860 occasions as per report of Google Scholar amongst which 22 h indexed. His area of research interest includes Power System analysis, Optimization, Distribution System, Restructuring and Smart Grid, Distribution planning, analysis and automation, optimum operation and planning of Power System. AI applications Deregulation Development of an OPF based Power System Simulator. He has guided a good number of Doctoral students for the award of Ph.D. degree. He has several IEEE transactions. Nandita Sanyal obtained her BE (Hons.) in Electrical Engineering, ME Electrical in Measurement and instrumentation and Ph.D. in Engg in 1993, 2003 and 2015 respectively from Jadavpur University. Her Research topic is Development of Image processing algorithm using Bacterial Foraging Optimization. She is presently Head of the Department of Electrical Engineering B.P. Poddar Institute of Management and Technology Kolkata. She has few International Journal Publication in Elsevier and has a chapter in book of Computational Intelligence in Image processing Applications by Springer Germany. Nandita worked in Swedish Multinational ESAB India Limited for eight years. Where she was engaged in design and development of Welding Transformers and Rectifiers. She does consultancy in small scale industries. She is Executive Committee member of IEEE CSS IMS joint Chapter Kolkata Section.

Control Systems

Anti-windup Control of Nonlinear Cascade Systems with Particle Swarm Optimization Parameter Tuning Fernando Serrano and Josep M. Rossell

Abstract Assuming that many physical models can be decoupled, an anti-windup control scheme for nonlinear cascade systems is proposed. Taking into account that saturation appears frequently, in order to overcome this difficulty, an efficient control approach is developed. The paper is divided into two parts; First, the design of a cascade control system with dynamic controllers in the inner and outer loops, considering the closed-loop stability in the controller design with a suitable antiwindup compensator; Secondly, a PID cascade controller design in the inner and outer loop is presented, when the parameter tuning in both control schemes is done by particle swarm optimization (PSO). However, in this case, the implementation of an anti-windup compensator is not needed. Apart from the theoretical background, two numerical examples are shown to corroborate the provided results.

1 Introduction Cascade control systems have been investigated since several decades. In the SISO linear case, as it is known, the controllers are tuned in sequence, first by tuning the inner loop and then the outer loop. Usually, the kind of controllers implemented are proportional–integral–derivative (PID). In recent years, the research about cascade control systems has been extended to the nonlinear case, considering that many physical systems such as mechanical, electrical, power systems, and chemical systems can be controlled and stabilized by means of this approach. The design is possiThis work was partially supported by the Spanish Ministry of Economy and Competitiveness under Grant DPI2015-64170-R(MINECO/FEDER). F. Serrano Central American Technical University (UNITEC), Zona Jacaleapa, Tegucigalpa, Honduras e-mail: [email protected] J. M. Rossell (B) Department of Mathematics, Univ. Politècnica de Catalunya (UPC), Avda. Bases de Manresa 61-73, 08242 Manresa, Spain e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_1

3

4

F. Serrano and J. M. Rossell

ble because a decoupled system can be divided into an inner and an outer loop, improving the performance in comparison with single loop control techniques. In the literature, the research about this topic is limited but an example can be found in [1] where a cascade control system is designed for the stabilization of underactuated mechanical systems. Although the anti-windup control problem for cascade control systems has not been investigated extensively, there are interesting results in single loop anti-windup design. In [2], an anti-windup control design is developed for the control of Takagi–Sugeno systems and a reliable state feedback control of Takagi— Sugeno fuzzy systems with sensor faults can be seen in [3]. A control scheme for disturbance observer systems is provided in [4], dealing with the saturation torque. Other theoretical and applied studies have been presented in [5], where the results are implemented in single-loop linear systems and the gain matrices are computed by using linear matrix inequalities (LMIs). Based on a linear approach, an anti-windup control scheme for an underwater vehicle is given in [6] and an anti-windup approach for nonlinear systems can be found in [7]. Other interesting works related to this topic are given in [8–10]. In this paper, an anti-windup control scheme is proposed for the stabilization of cascade nonlinear systems, which is developed in two parts. The first one is a dynamic controller implemented in the inner and outer loop. The closed-loop stability of the system is based on the theory stability of Lyapunov [11]. An anti-windup compensator is designed in order to reduce the unwanted effects of windup such as poor performance or even instability. The second part is done by implementing PID controllers in the inner and outer loop but now without anti-windup compensation. In the first and second part of this study, the gain matrices are tuned by particle swarm optimization [12–15]. The paper is organized as follows: In Sect. 2, the design of an anti-windup control scheme for cascade control systems, implementing dynamic controllers in the inner and outer loop, is developed. In Sect. 3, a PID cascade control system design is presented by considering input saturation but without the anti-windup compensator. In Sect. 4, a PSO algorithm is supplied in order to tune the gain matrices for both approaches. Two numerical examples are given in Sect. 5 and the conclusions can be found in Sect. 6.

2 Anti-windup Cascade Dynamic Controller Design This section is devoted to designing an anti-windup controller for nonlinear cascade systems. This strategy implements dynamic controllers in the inner and outer loops with gain matrices that help to improve the system performance. The controllers are tuned, as explained in Sect. 4, by a particle swarm optimization algorithm to reduce the integral square error, i.e., the difference between the reference variable and the output of the outer system. The same applies to the inner system. The main idea of this first approach is to design an appropriate anti-windup compensator to deal with the unwanted effects when saturation appears in the inner loop. Even when the gain

Anti-windup Control of Nonlinear Cascade Systems … r1 Controller 2

xc2 = yc2 r2

Controller 1

xc1 = u

Saturation

+ uc2

xc2 uT [u−φ(u)]

φ(u)

yc

5

System 1

System 2

− u−φ(u)

y1 = x1 = uc

y2 = x2 = uc2

Anti-windup Compensator

Fig. 1 Cascade anti-windup dynamic control scheme

matrices are tuned by a PSO algorithm, the closed-loop stability of the inner and the overall systems are proved by the method of Lyapunov [11]. The obtained results are compared with the approaches given in [16, 17]. A. Inner loop dynamic controller design Consider the cascade control dynamic scheme shown in Fig. 1. The inner loop system (system 1) is formed by x˙ 1 (t) = −Ax1 (t) + f1 (x1 (t)) + ϕ(u(t)), y1 (t) = x1 (t) = uc (t),

(1)

where A ∈ n×n is an appropriate positive definite matrix due to the inner loop system being minimum phase in order to facilitate the particle swarm optimization parameter tuning; x1 (t) ∈ n is the state vector; f 1 (x 1 (t)) is a nonlinear vector function; ϕ(·) is the saturation nonlinearity; u(t) ∈ n is the input vector; y1 (t) ∈ n is the controller output; and uc (t) ∈ n is the controller input vector. The inner loop controller is given by x˙ c1 (t) = −Kxc1 (t) + r2 (t) − [u(t) − ϕ(u(t))] + uc (t), yc (t) = xc1 (t) = u(t),

(2)

where K ∈ n×n is a positive definite gain matrix with a negative sign to make the closed-loop system of minimum phase type; xc1 (t) ∈ n is the controller state vector; r2 (t) ∈ n , the reference vector; and u(t)—ϕ(u(t)) is the compensation term. Before proving the closed-loop stability of the inner loop, the definition of sector condition is needed [1]. Definition 1 The sector condition for the saturation nonlinearity is given by uT (t)[u(t) − ϕ(u(t))] > 0.

(3)

6

F. Serrano and J. M. Rossell

In the following theorem, the closed-loop stability of the inner loop is proved in order to obtain a stable controller by considering that the gain matrix K is tuned by means of a PSO algorithm, which is detailed in Sect. 4. Theorem 1 There exists a gain matrix K for the controller (2) such that the closedloop of the inner loop is stable. Proof Consider the following storage function [11]:     V x1 (t), xc1 (t) = Vs (x1 (t)) + σ Vsc xc1 (t) ,

(4)

with σ > 0 and where 1 T x (t)x1 (t), 2 1   1 Vsc xc1 (t) = xcT1 (t)xc1 (t), 2

Vs (x1 (t)) =

(5)

and the auxiliary input variable υ(t) = −f1 (x1 (t)) − ϕ(u(t)).

(6)

Then, the system (1) becomes x˙ 1 (t) = −Ax1 (t) − v(t)

(7)

and defining the input variable w(t) = −r2 (t) − uc (t),

(8)

the system (2) can be written as x˙ c1 (t) = −Kxc1 (t) − w(t) − [u(t) − ϕ(u(t))].

(9)

Now, obtaining the first derivative of (4) along with (7) and (9) yields   V˙ x1 (t), xc1 (t) = −x1T (t)Ax1 (t) − x1T (t)v(t)   − σ xcT1 (t)Kxc1 (t) + xcT1 (t)w(t) + xcT1 (t)[u(t) − ϕ(u(t))] . (10) From (1) and (2), x 1 (t) = y1 (t) and xc1 (t) = u(t). Then, the Eq. (10) becomes   V˙ x1 (t), xc1 (t) = −y1T (t)Ay1 (t) − y1T (t)v(t)   − σ ycT (t)Kyc (t) + ycT (t)w(t) + uT (t)[u(t) − ϕ(u(t))]

(11)

and using Definition 1 in (11) and considering that the system is zero state observable [7], i.e., yi (t) = 0 implies uci (t) = 0 and x i (t) = 0, for i = 1, 2, we obtain

Anti-windup Control of Nonlinear Cascade Systems …

  V˙ x1 (t), xc1 (t) ≤ 0

7

(12)

and the theorem is proved [11]. B. Outer loop and overall dynamic controller design Before deriving the outer loop controller and proving the overall closed-loop stability, it is necessary to make the following change of variables  T x¯ (t) = x1T (t), xcT1 (t) ,  T u¯ (t) = uT (t), ϕ T (u(t)), r2T (t), ucT (t) .

(13)

The equivalent closed-loop system is ¯ x(t) + f¯ (¯x(t)) + B¯ ¯ u(t), x˙¯ (t) = A¯

(14)

with     f (x (t)) −A 0n , , f¯ (¯x(t)) = 1 1 A¯ = 0 0n −K   ¯B = 0n In 0n 0n , −In In In In

(15)

where I n and 0n are the identity and the zero matrix of appropriate dimensions, respectively. Now, consider the outer loop dynamic equation (system 2) given by x˙ 2 (t) = −A2 x2 (t) + f2 (x2 (t)) + u2 (t),

(16)

where A2 ∈ m×m is a positive definite matrix; x2 (t) ∈ m is the state vector; f 2 (x 2 (t)), the nonlinearity vector; and u2 (t) ∈ m is the input vector. The overall closed-loop system is obtained by selecting the augmented vectors T  x˜ (t) = x1T (t), xcT1 (t), x2T (t) ,  T u˜ (t) = uT (t), ϕ T (u(t)), r2T (t), ucT (t), u2T (t) .

(17)

˜ x(t) + f˜ (˜x(t)) + B˜ ˜ u(t) x˙˜ (t) = A˜ y˜ (t) = x2 (t),

(18)

Then,

where the output is given by

8

F. Serrano and J. M. Rossell

⎡ ⎤  x1 (t) y˜ (t) = 0n 0n Im ⎣ xc1 (t) ⎦ = x2 (t) x2 (t) 

(19)

with x2 (t) = uc2 (t) ∈ m the controller input, ⎡

⎤ ⎡ ⎤ −A 0n 0m f1 (x1 (t)) ⎦, A˜ = ⎣ 0n −K 0m ⎦, f˜ (˜x(t)) = ⎣ 0 f2 (x2 (t)) 0n 0n −A2 ⎤ ⎡ 0n In 0n 0n 0m B˜ = ⎣ −In In In In 0m ⎦ 0n 0n 0n 0n Im

(20)

and rewriting ˜ x(t) − v˜2 (t) x˙˜ (t) = A˜

(21)

˜ u(t). v˜2 (t) = −f˜ (˜x(t)) − B˜

(22)

with

Consider the following outer loop dynamic controller: x˙ c2 (t) = −K2 xc2 (t) − r1 (t) − xc2 (t)˜z T (t)[u(t) − ϕ(u(t))] − uc2 (t) = −K2 xc2 (t) − v2 (t) − xc2 (t)˜z T (t)[u(t) − ϕ(u(t))],

(23)

where xc2 (t) ∈ m is the dynamic controller state vector, K2 ∈ m×m is the controller gain matrix, r1 (t) ∈ m is the reference vector and defining v2 (t) = r1 (t) + uc2 (t), yc2 (t) = xc2 (t),

(24)

where yc2 (t) ∈ m is the controller output, together with an extra output ⎡ ⎤  x1 (t) z˜ (t) = 0n In 0m ⎣ xc1 (t) ⎦ = xc1 (t). x2 (t) 

(25)

Then, the following theorem can be stated. Theorem 2 There exists a gain matrix K for the controller (23) such that the closedloop of the outer loop (overall system) is stable. Proof Consider the storage function for the overall closed-loop system

Anti-windup Control of Nonlinear Cascade Systems …

9

    V x˜ (t), xc2 (t) = Vs (˜x(t)) + σ Vsc xc2 (t) ,

(26)

with σ > 0 and where 1 T x˜ (t)˜x(t), 2   1 Vsc xc2 (t) = xcT2 (t)xc2 (t). 2

Vs (˜x(t)) =

(27)

Now, deriving (26) along the trajectory given in (21) and (23), we obtain   ˜ x(t) − x˜ T (t)v˜2 (t) V˙ x˜ (t), xc2 (t) = x˜ T (t)A˜   − σ xcT2 (t) K2 xc2 (t) + v2 (t) + xc2 (t)uT (t)[u(t) − ϕ(u(t))] . (28) Then, from Definition 1, and considering that the system is zero state observable,   V˙ x˜ (t), xc2 (t) ≤ 0

(29)

is obtained and the closed-loop stability of the outer loop is ensured.

3 PID Cascade Control System Design A PID cascade control system is formed by two parts: An inner loop PID controller [13] and an outer loop PID controller (see Fig. 2). In this case, it is not necessary to implement an anti-windup compensator because the gain matrices are tuned by the particle swarm optimization routine shown in the next section. The PID controllers for the inner and outer loop are given by the following equations: t ec1 (t)dt + Kd1 e˙ c1 (t), c1 (t) = Kp1 ec1 (t) + Ki1 0

r1

+ ec1

c1 PID Controller 2



c2

PID Controller 1

ec2

+



Fig. 2 Cascade PID control scheme

Saturation System 1

y1 = x1

System 2

y2 = x2

10

F. Serrano and J. M. Rossell



t

c2 (t) = Kp2 ec2 (t) + Ki2

ec2 (t)dt + Kd2 e˙ c2 (t),

(30)

0

where c1 (t) ∈ n is the output controller 1; c2 (t) ∈ m , the output controller 2; ec1 (t) = r1 (t) − y2 (t), the error variable for the controller 1; ec2 (t) = c2 (t) − y1 (t), the error variable for the controller 2; Kp1 , Ki1 , Kd1 ∈ n are the proportional, integral and derivative gain matrices, respectively, for the controller 1 and Kp2 , Ki2 , Kd2 ∈ m for the controller 2.

4 Particle Swarm Optimization Routine for Cascade Anti-windup Controller Design Particle swarm optimization routines have been recently implemented for the parameter tuning for PIDs and other controllers [12–15, 18]. In this study, the first step is to determine an anti-windup scheme for a cascade dynamic controller in order to find an appropriate controller and compensator ensuring the closed-loop overall stability. The gain matrices are found by using a PSO algorithm to minimize the integral squared error of the overall system [18] where Fi (ei ) =

m

ei2 (j) j

(31)

j=1

is the objective function that minimizes the error ei (n) = r1 (n) − y2 (n) with the time difference j . The PSO algorithm that allows us to find the gain matrices K, K 2 for the dynamic controller and Kp1 , Ki1 , Kd1 , Kp2 , Ki2 , Kd2 for the PID controller scheme is given by while (gbest > r 1 and gbest < r 2 … gbest < r n and j < 100000) for (int i = 0; i < paramnum; i ++) V[i] = V[i] + c1 (rand()) (pbest[i] – c2 (rand()))(gbest – X[i]); X[i] = X[i] + V[i]; F = Objectivefunction(X) if (F[i] ≤ Fpbest[i]) pbest[i] = X[i]; Fpbest[i] = F[i]; if (F[1] ≤ Fgbest) gbest = X[i]; Fgbest = F[1]; where X is the particle position for the gain matrices component; V is the particle velocity; pbest and gbest are the best particle positions; and Fgbest is the final result obtained by the objective function.

Anti-windup Control of Nonlinear Cascade Systems …

11

5 Numerical Examples The following systems are used in two examples: 3 2 (t) + x11 (t)x12 (t) + u1 (t), x˙ 11 (t) = −x11 (t) − x12 (t) + x11 4 2 x˙ 12 (t) = −x12 (t) + 2x11 (t) + x12 (t)x11 (t) + u2 (t),

(32)

3 x˙ 21 (t) = x21 (t) + u1 (t), 3 x˙ 22 (t) = x22 (t) − x21 (t)x22 (t) + u2 (t),

(33)

where (32) and (33) will be the systems (1) and (2) for the Example 1, and 2 (t) + u1 (t), x˙ 11 (t) = −x11 (t) − 0.0001x11 2 x˙ 12 (t) = −x12 (t) − 0.0001x12 (t) + u2 (t),

(34)

2 x˙ 21 (t) = −x21 (t) − 0.0001x21 (t) + u1 (t), 2 x˙ 22 (t) = −x22 (t) − 0.0001x22 (t) + u2 (t),

⎧ x ≤ −19800 ⎨ −19800 for ϕ(x) = x for −19800 < x < 19800 , ⎩ 19800 for x ≥ 19800

(35)

(36)

where (34) and (35) will be the systems (1) and (2) for Example 2, with the saturation function given in (36). A. Example 1: Dynamic controller experiment In this subsection, a numerical example to test the dynamic controller design is shown. The obtained results are compared with the results presented in [16, 17], considering that the strategies evinced in both studies are used in a cascade controller configuration. Figures 3 and 4 depict the trajectories obtained by the variables x 11 (t) and x 21 (t), where the latest variable is the output of the overall closed-loop system when a reference is used to reach the origin or the equilibrium point. Note that these variables reach more efficiently the equilibrium point in comparison with the strategies given in [16, 17], with less overshoot and faster response. The same occurs for the variable x 22 (t) shown in Fig. 5. In Figs. 6 and 7, the controller inputs xc11 (t) and xc22 (t) are presented and the control effort generated by the control strategy is also smaller than that obtained in [16, 17].

12

F. Serrano and J. M. Rossell

Fig. 3 Variable x 11 (t)

0.01

Proposed Controller Kanamori Oliveira et al

Variable x11

0.005

0

-0.005

-0.01

0

50

100

150

200

250

300

Time (s)

Fig. 4 Variable x 21 (t)

0.15

Proposed Controller Kanamori Oliveira et al

Variable x21

0.1

0.05

0

-0.05

-0.1

0

50

100

150

200

250

300

Time (s)

Fig. 5 Variable x 22 (t)

0.02 Proposed Controller Kanamori Oliveira et al

Variable x22

0.01

0

-0.01

-0.02

-0.03

0

50

100

150

Time (s)

200

250

300

Anti-windup Control of Nonlinear Cascade Systems … Fig. 6 Control input xc11 (t)

13

0.02

Proposed Controller Kanamori Oliveira et al

Variable xc11

0.01

0

-0.01

-0.02

-0.03

0

50

100

150

200

250

300

Time (s)

Fig. 7 Control input xc22 (t)

0.06

Proposed Controller Kanamori Oliveira et al

Variable xc22

0.04

0.02

0

-0.02

-0.04

-0.06

0

50

100

150

200

250

300

Time (s)

B. Example 2: PID controller experiment In this example, the PID control gain matrices are tuned by a PSO algorithm and the obtained results are compared with the approaches given in [19, 20], with the origin as the equilibrium point. The control approaches in [19, 20] have been modified to operate in a cascade closed-loop configuration. The variable x 22 (t) is depicted in Fig. 8 and the desired final value has fewer oscillations and faster response in our case. Finally, in Figs. 9 and 10, the respective PID controller outputs c11 (t) and c22 (t) are shown and the control effort is smaller and with fewer oscillations, even when saturation is found in the input.

14

F. Serrano and J. M. Rossell

Fig. 8 Variable x 22 (t)

100

Proposed Strategy Tahoun Huang et al

80

Variable X22

60

40

20

0

-20

0

1

2

3

4

5

6

7

Time (s)

Fig. 9 Control variable c11 (t)

200

Proposed Strategy Tahoun Huang et al

Variable C11

100

0

-100

-200

0

1

2

3

4

5

6

7

Time (s)

Fig. 10 Control variable c22 (t)

0

Proposed Strategy Tahoun Huang et al

-5

Variable C22

-10

-15

-20

-25

-30

0

1

2

3

4

Time (s)

5

6

7

Anti-windup Control of Nonlinear Cascade Systems …

15

6 Conclusions Two anti-windup schemes for nonlinear cascade systems have been proposed and, when input saturation appears, the system performance is improved. The results represent a contribution in some physical systems such as mechanical, aeronautical, electrical, power, and energy systems, considering that saturation is a common phenomenon affecting them.

References 1. Mehdi N, Rehan M, Malik FM, Bhatti AI, Tufail M (2014) A novel anti-windup framework for cascade control systems: an application to underactuated mechanical systems. ISA Trans 53(3):802–815 2. Nguyen A, Dequidt A, Dambrine M (2015) Anti-windup based dynamic output feedback controller design with performance consideration for constrained Takagi Sugeno systems. Eng Appl Artif Intell 40:76–83 3. Dong J, Yang G-H (2015) Reliable state feedback control of T– S fuzzy systems with sensor faults. IEEE Trans Fuzzy Syst 23(2):421–433 4. Gao X, Komada S, Hori T (1999) A wind-up restraint control of disturbance observer system for saturation of actuator torque. In: IEEE international conference on systems, man, and cybernetics, vol 1, pp 84–88 5. Silva JD, Tarbouriech S (2003) Anti-windup design with guaranteed regions of stability: an LMI based approach. In: Proceedings of the 42nd IEEE conference on decision and control, vol 5, pp 4451–4456 6. Folcher JP (2004) LMI based anti-windup control for an underwater robot with propellers saturations. In: Proceedings of the IEEE international conference on control applications, vol 1, pp 32–37 7. Oliveira MZ, Da Silva JMG, Coutinho D, Tarbouriech S (2011) Anti-windup design for a class of multivariable nonlinear control systems: an LMI based approach. In: 50th IEEE conference on decision and control and european control conference, pp 4797–4802 8. Zhai D, An L, Li J, Zhang Q (2016) Fault detection for stochastic parameter-varying Markovian jump systems with application to networked control systems. Appl Math Model 40(3):2368–2383 9. Zhai D, An L, Li J, Zhang Q (2016) Adaptive fuzzy fault-tolerant control with guaranteed tracking performance for nonlinear strict-feedback systems. Fuzzy Set Syst 302:80–100 10. Zhai D, Lu A-Y, Li J-H, Zhang Q-L (2016) Simultaneous fault detection and control for switched linear systems with mode-dependent average dwell-time. Appl Math Comput 273:767–792 11. Haddad W, Chellaboina V (2008) Nonlinear dynamical systems and control: a Lyapunov based approach. Princeton Press, Princeton 12. Gao H, Xu W (2011) Particle swarm algorithm with hybrid mutation strategy. Appl Soft Comput 11(8):5129–5142 13. Menhas MI, Wang L, Fei M, Pan H (2012) Comparative performance analysis of various binary coded PSO algorithms in multi-variable PID controller design. Exp Syst Appl 39(4):4390–4401 14. Wang L, Fu X, Mao Y, Menhas MI, Fei M (2012) A novel modified binary differential evolution algorithm and its applications. Neurocomputing 98:55–75 15. Wang Y, Li B, Weise T, Wang J, Yuan B, Tian Q (2011) Self-adaptive learning based particle swarm optimization. Inf Sci 181(20):4515–4538 16. Kanamori M (2012) Anti-windup adaptive law for Euler-Lagrange systems with actuator saturation. IFAC Proc 45(22):875–880

16

F. Serrano and J. M. Rossell

17. Oliveira MZ, Gomes da Silva JM, Coutinho DF, Tarbouriech S (2011) Anti-windup design for a class of nonlinear control systems. IFAC Proc Vol 44(1):13432–13437 18. Serrano FE, Flores MA (2015) C ++ library for fuzzy type-2 controller design with particle swarm optimization tuning. In: IEEE CONCAPAN 2015. Tegucigalpa, Honduras 19. Tahoun AH (2017) Anti-windup adaptive PID control design for a class of uncertain chaotic systems with input saturation. ISA Trans 66:176–184 20. Huang CQ, Peng XF, Wang JP (2008) Robust nonlinear PID controllers for anti-windup design of robot manipulators with an uncertain Jacobian matrix. Acta Autom Sin 34(9):1113–1121

Pollutant Profile Estimation Using Unscented Kalman Filter S. Metia, S. D. Oduro and A. P. Sinha

Abstract In this paper, we develop an estimation model for carbon monoxide (CO) air pollution concentrations. CO is an important pollutant which is used to calculate an air quality index (AQI). AQI becomes less reliable as the proportion of data missing due to equipment failure and periods of calibration increases. This paper presents the Unscented Kalman filter (UKF) to predict missing data of atmospheric carbon monoxide concentrations using the time series data of monitoring stations. Keywords Carbon monoxide (CO) · Unscented Kalman filter (UKF) · Air quality index (AQI)

1 Introduction Carbon monoxide (CO) is a product of incomplete combustion of fossil fuel, biofuel, and biomass burning [1] and oxidation of hydrocarbon compounds [2]. Exposure to low concentrations (e.g., from less than 10 ppm) of CO can affect organ systems. It is suggested that continuous exposure to CO may produce mild neurological effects [3]. Higher ambient CO concentrations are prone to heart disease, and low levels of CO adversely affect patients with heart disease when exercising [4]. Occupationally, exposure to CO is a major hazard to those dealing with the combustion of fuel. For example, fire fighters may be exposed to CO concentrations as high as 3,000 ppm [5]. The spatial and temporal variations of CO have been characterized S. Metia (B) Faculty of Engineering and IT, University of Technology Sydney, Sydney, Australia e-mail: [email protected] S. D. Oduro Department of Mechanical and Automotive Technology Education College of Technology Education Kumasi, University of Education Winneba, Kumasi, Ghana e-mail: [email protected] A. P. Sinha (B) Department of Electronics and Communication Engineering, BIT Sindri, Dhanbad, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_2

17

18

S. Metia et al.

based on the measurements using different techniques such as remote sensing from space, ground-based remote sensing, and in situ sampling. In [6], the authors have reviewed the concentration levels of carbon dioxide (CO2), carbon monoxide (CO), particulate matter (PM), metal elements, nitrogen oxides (NOx), ozone (O3 ), sulfur dioxide (SO2 ), volatile organic compounds (VOCs), polycyclic aromatic hydrocarbons (PAHs), and persistent organic pollutants (POPs). In [7], the authors have used the modeling system which is based on an adaptive nonlinear state-space-based filter. The filtering equations have been solved using an Extended Kalman filter. In [8], authors showed how temporal (i.e., time series) Gaussian process regression models in machine learning could be reformulated as linear Gaussian state-space models, which could be solved exactly with classical Kalman filtering theory. They reformulated the model and showed Matérn family of covariance functions analytically and for squared exponential covariance function by applying spectral Taylor series approximation. The EKF maintains the classical and mathematically efficient repetitive update form of the Kalman filter (KF); the first-order Taylor series approximation works well for nonlinear functions [9] when the nonlinear function is “mild” in nature. It is suboptimal and can easily lead to deviation. The linearized transformations are based on a linear function which is used to calculate the error propagation function. The linearization calculation is valid when the Jacobian matrix exists. The calculation of Jacobian matrix is difficult and sometimes it leads to error-prone results. To overcome the deficiencies of linearization, the unscented transformation (UT) [10, 11] was proposed. It is based on direct and accurate mathematical calculation for transforming mean and covariance information. Based on UT, Julier et al. [11, 12] developed the Unscented Kalman filter (UKF) as a derivative-free alternative to EKF in the framework of state estimation. The paper is arranged as follows: Sect. 2 presents the methodology. Section 3 discusses the relationship between Matérn covariance function with the Kalman Filter. Section 4 introduces the Unscented Kalman Filter. In Sect. 5, we discuss the evaluation and analysis of the UKF. Finally, Sect. 6 summarizes the paper with some concluding remarks.

2 Methodology The observation vector is the observed pollutant concentrations of CO at the surface monitoring stations over New South Wales (NSW). The relationship between observation vector y with the noise and state vector x can be described as follows under the forward model: y = F(x, b) + ψ,

(1)

where b is the model parameter vector, ψ is the noise vector, and F stands for the forward model. Figure 1 shows the forward model using Matérn function based UKF.

Pollutant Profile Estimation Using Unscented …

19

Fig. 1 Flowchart of estimating station data using Matérn function based UKF using the forward model

3 Matérn Covariance Function Two points are separated by τ distance, where the Matérn covariance function is given by [13, 14] 21−ν kν = σ Γ (ν) 2

ν  √  √ 2ν 2ν τ Kν τ , l l

(2)

where Γ is the gamma function, K ν is the modified Bessel function of the second kind, and l and σ are nonnegative parameters of the covariance. Consider the air pollutant dispersion model and Matérn covariance function as described in ⎡ ⎤ ⎡ ⎤ 0 1 0 0 d x(t) ⎣ = 0 0 1 ⎦ x(t) + ⎣ 0 ⎦q(t) dt −λ3 −3λ2 −3λ 1  y(t) = 1 0 0 x(t) + r (t),

(3)

where x(t) is the state vector, q(t) is white noise, r (t) is measurement noise, and λ is a coefficient depending on the correlation length and smoothness of the process.

4 Unscented Kalman Filter A discrete-time nonlinear system can be described as xk = f k (xk−1 , u k−1 ) + qk−1 yk = h k (xk , u k ) + rk ,

(4)

20

S. Metia et al.

where xk ∈ Rn is the state, yk ∈ Rm is the measurement, u k ∈ Rv is the input, qk−1 ∈ Rn is a Gaussian process noise qk−1 ∼ N (0, Q k−1 ), rk ∈ Rm is a Gaussian measurement noise rk ∼ N (0, Rk ), and Q k−1 and Rk are covariances of qk−1 and rk . The generic UKF is summarized as follows: Step (1) For the system state x, initialize with xˆ0 = E[x0 ] P0 = E[(x0 − xˆ0 )(x0 − xˆ0 )T ],

(5)

where x0 , xˆ0 , and P0 represent the initial state vectors, predicted values, and covariance, respectively. Step (2) Sigma point calculation based on the (2N + 1)dimensional random variable xk−1 with mean xˆk−1 and covariance Pˆk−1 is approximated by the sigma points selected in the following equations: X 0,k−1 = xˆk−1



X i,k−1 = xˆk−1 + γ

X i,k−1 = xˆk−1 − γ

Pˆk−1 , i = 1, 2, . . . , L

(6)

Pˆk−1 , i = L + 1, L + 2, . . . , 2L

√ where γ = L + λ, λ = α 2 (L + κ) − L, L = 2N + 1, and both λ and γ are scaling parameters. The constant α determines the spread of the sigma points around xˆ and is usually set to 0.0001 ≤ α ≤ 1, and the parameter κ is used to control the covariance matrix P. Step (3) Each sigma point is used to obtain a transformed process model as X k/k−1 = ψ xˆk xˆk¯ − γ Pk xˆk− + γ Pk .

(7)

The predicted mean and covariances of the signal state are obtained from the following equations: xˆ k¯ =

2L 

wi(m) X i,k/k−1

i=0

Pk− =

2L 

  T wi(c) X i,k/k−1 − xˆk− X i,k/k−1 − xˆk− + Q k .

i=0

The weighting parameters wi(m) and wi(c) are given by w0(c) =

 λ λ  w0(m) = 1 − α2 + β L +λ L +λ

(8)

Pollutant Profile Estimation Using Unscented …

wi(m) = wi(c) =

λ , i = 1, 2, 3, . . . , 2L . 2(L + λ)

21

(9)

Here, β is used to incorporate prior knowledge of the distribution of x. The mean and covariance of yk are approximated using the weighted sample mean and covariance of the posterior sigma points as yˆk¯ =

2L 

wi(m) Yi,k/k−1

i=0

  Yk/k−1 = h X k/k−1 , i = 0, 1, . . . , 2L Pyk yk =

2L 

(10)

  T wi(c) Yi,k/k−1 − yˆk Yi,k/k−1 − yˆk + Rk

i=0

Pxk yk =

2L 

  T wi(c) X i,k/k−1 − yˆk Yi,k/k−1 − yˆk .

(11)

i=0

The Kalman gain, along with the signal state and error covariance updates, is obtained as K k = Pxk yk Py−1 k yk   xˆk = xˆk¯ + K k yk − yˆk¯

(12)

− Pk = Pk−1 − K k Pyk yk K kT .

The values of the system state xˆ and covariance matrix Pk become the input of the successive prediction correction loop. Through a proper choice of the sigma points, i.e., the parameters α, λ, and β and the covariance matrices Q 0 and R, the UKF assures a better performance than the EKF in estimating fundamental components of a signal buried in noise.

5 Results and Discussion The UKF performance is promising compared to the EKF estimation in the time series domain where the UKF prediction is more accurate than the EKF estimation. Figure 2 shows the profile of CO at Chullora station by monitoring as well as using the EKF and the UKF. CO concentration is measured in parts per million (ppm). The UKF and the EKF are implemented to estimate the time series profile of CO pollutant at Chullora station. Both the estimation time series show that both the profiles follow

22

S. Metia et al.

Fig. 2 CO level at Chullora station from February 1 to 6, 2018

Table 1 Summary statistics for CO pollutant profile

Filter

R2

MSE

p-value

EKF

0.6437

0.00320

2.69 × 10−10

UKF

0.9868

0.00011

1.09 × 10−10

the observation profile of CO at Chullora station. It is elucidated the UKF estimation of CO is more accurate than the EKF estimation. The MSE of CO profile is shown in Table 1. The EKF shows 0.00320 as MSE and the UKF shows 0.00011 as MSE. From Table 1, it is interpreted that the UKF is more accurate than the EKF. The UKF can handle uncertainties better than the EKF in terms of the pollutant profile estimation. As shown in Figs. 3 and 4, the regression plots are plotted for the EKF prediction and the UKF prediction, respectively. These

Fig. 3 EKF prediction correlation coefficient for CO at Chullora station

Pollutant Profile Estimation Using Unscented …

23

Fig. 4 UKF prediction correlation coefficient for CO at Chullora station

regression plots are plotted between the output and target. The correlation between the measurement data collected at the monitoring station Chullora and the UKF estimation is highly correlated. It shows that the UKF is more accurate than the EKF estimation scheme. Table 2 shows CO concentration at different monitoring stations across New South Wales. 33 monitoring stations are measuring different pollutants round the clock. 11 Table 2 CO concentration on February 7, 2018 at 6 A.M.

Station

Monitoring data (ppm)

EKF data (ppm)

UKF data (ppm)

Rozelle

0.1

0.1165

0.0982

Liverpool

0.5

0.4309

0.4993

0.3979

0.5060

Chullora Wyong

0.1

0.0945

0.1000

Newcastle

0.4

0.3037

0.3991

Wollongong

0.2

0.1822

0.2107

Parramatta North

0.4

0.2825

0.3851

Prospect

0.2

0.1791

0.1894

Campbelltown 0.3

0.2813

0.3087

Camden

0.1

0.0984

0.1000

Macquarie Park

0.3

0.2956

0.3050

24

S. Metia et al.

monitoring stations have the facility to measure CO concentration. CO data is missing at Chullora station on February 7, 2018, at 6 A.M. All monitoring data are estimated by using the EKF and the UKF. Figure 5 shows the spatial plot of CO using station data. From the climatic point of view, it can also be interpreted from Fig. 5 that the CO is produced by motor vehicle exhaust and power plants. Sydney Central Business District (CBD) and surrounding areas have a high concentration of CO. Beresfield and surrounding areas have coal-based power plants. That is the main reason, CO concentration is higher in the right corner of the map. A similar trend is shown in Figs. 6 and 7. The EKF estimated data are used in Fig. 6 and the UKF estimated data are used in Fig. 7. Figure 8 shows the difference between monitoring station data and the EKF data as well as Fig. 9 shows the difference between monitoring data and the UKF data. It shows that the UKF has more accurate estimation than the EKF.

Fig. 5 CO distribution on February 7, 2018 at 6 A.M., based on monitoring station data

Pollutant Profile Estimation Using Unscented … Fig. 6 CO distribution on February 7, 2018 at 6 A.M., based on EKF data

Fig. 7 CO distribution on February 7, 2018 at 6 A.M., based on UKF data

25

26 Fig. 8 CO difference between monitoring station data and EKF data

Fig. 9 CO difference between monitoring station data and UKF data

S. Metia et al.

Pollutant Profile Estimation Using Unscented …

27

6 Conclusion Ambient CO is a part of air quality. Sydney is a metropolitan city where air quality monitoring is a part of the local government policy. For accurate air quality monitoring, the mathematical model of air quality should be accurate. We have proposed the UKF as an estimation and prediction tools to estimate the profile of CO in time and spatial domains. We have presented the design of UKF which is based on Mat`ern covariance function. The UKF is used to estimate the missing profile of CO pollutant in time series or spatial domains. The procedure shows high efficiency of the estimation of CO pollutant profile in time as well as spatial domains. The UKF is robust which can handle system process and measurement noises efficiently. In terms of smoothing, the Mat`ern covariance function-based UKF helps smoothing the profile of pollutant in time and spatial domains. It can be seen in the results of the comparison table where measured station data are compared with estimation data across New South Wales.

References 1. Zhou Y, Mao H, Demerjian K, Hogrefe C, Liu J (2017) Regional and hemispheric influences on temporal variability in baseline carbon monoxide and ozone over the Northeast US. Atmos Environ 164:309–324 2. Gaeggelera K, Prevota ASH, Dommena J, Legreid G, Reimann S, Baltensperger U (2008) Residential wood burning in an Alpine valley as a source for oxygenated volatile organic compounds, hydrocarbons and organic acids. Atmos Environ 42(35):8278–8287 3. Spinazze A, Cattaneo A, Garramone G, Cavallo DM (2013) Temporal variation of sizefractionated particulate matter and carbon monoxide in selected microenvironments of the milan urban area. J Occup Environ Hyg 10(11):652–662 4. Dahms TE, Younis LT, Wiens RD, Zarnegar S, Byers SL, Chaitman BR (1993) Effects of carbon monoxide exposure in patients with documented cardiac arrhythmias. J Am Coll Cardiol 21(2):442–450 5. Mohajer NA, Zuidema C, Sousan S, Hallett L, Tatum M, Rule AM, Thomas G, Peters TM, Koehle K (2018) Evaluation of low-cost electro-chemical sensors for environmental monitoring of ozone, nitrogen dioxide, and carbon monoxide. J Occup Environ Hyg 15(2):87–98 6. Omidvarborna H, Baawain M, Al-Mamun A (2018) Ambient air quality and exposure assessment study of the Gulf Cooperation Council countries: a critical review. Sci Total Environ 636:437–448 7. Zolghadri A, Cazaurang F (2006) Adaptive nonlinear state-space modelling for the prediction of daily mean PM10 concentrations. Environ Model Softw 21(6):885–894 8. Hartikainen J, Särkkä S (2010) Kalman filtering and smoothing solutions to temporal Gaussian process regression models. In: 2010 IEEE international workshop on machine learning for signal processing, August 2010, pp. 379–384 9. Arasaratnam I, Haykin S (2009) Cubature Kalman filters. IEEE Trans Autom Control 54(6):1254–1269 10. Julier S, Uhlmann J, Durrant-Whyte HF (2000) A new method for the nonlinear transformation of means and covariances in filters and estimators. IEEE Trans Autom Control 45(3):477–482 11. Reddy JBV, Dash PK, Samantaray R, Moharana AK (2009) Fast tracking of power quality disturbance signals using an optimized unscented filter. IEEE Trans Instrum Meas 58(12):3943–3952

28

S. Metia et al.

12. Julier SJ, Uhlmann JK (2004) Unscented filtering and nonlinear estimation. Proc IEEE 92(3):401–422 13. Metia S, Oduro SD, Duc HN, Ha Q (2016) Inverse air-pollutant emission and prediction using extended fractional Kalman filtering. IEEE J Sel Top Appl Earth Obs Remote Sens 9(5):2051–2063 14. Metia S, Ha QP, Duc HN, Azzi M, Estimation of power plant emissions with unscented Kalman filter. IEEE J Sel Top Appl Earth Obs Remote Sens 11(8):2763–2772

Determination of Model Order of an Electrochemical System: A Case Study with Electronic Tongue Sanjeev Kumar and Arunangshu Ghosh

Abstract The paper presents a technique to determine the optimal model order of an electrochemical system with the help of system identification. The study has been performed for the case of different tea samples on a voltammetric electronic tongue. The transfer function model of the system with different combinations of number of poles and zeros are identified using the response data obtained from the electronic tongue. Based upon the normalized root mean square (NRMSE) criteria, the model fit for different model orders are compared and the optimal order of the system is determined. Keywords Electrochemical system · System identification · Electronic tongue · Sensor modeling

1 Introduction A model is a mathematical representation of the relation between the input and the output of a system that well signifies the dynamics of a system. The model of a system is defined by its structure and the corresponding parameters [1]. The model order is another important aspect to look for as it is directly linked with the model structure. The order of a system is defined by the highest value of exponent in the denominator polynomial when the system is represented by a proper transfer function. In typical control engineering words, the number of poles present in the transfer function determines the system order. In a practical system, the determination of the model order is an important task as it helps in understanding the system dynamics. An inappropriate model order may lead to improper interpretation of the model response when compared to the actual characteristics of the system. Sometimes a lower model order can perfectly define the dynamics of the system. The lower model order has very less mathematical S. Kumar (B) · A. Ghosh Department of Electrical Engineering, National Institute of Technology Patna, Ashok Rajpath, Patna 800005, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_3

29

30

S. Kumar and A. Ghosh

complexities. The time and frequency response of lower order models are also very clear and easy to understand. But there can be situation in which the system cannot be defined by lower orders. In those cases one has to switch to higher order models. However, the higher order systems have model complexities and they are difficult to analyze. An electrochemical system is the one which has a very complex behavior as the response of the system is due to the electrochemical reaction taking place at the electrode- electrolyte interface. A typical electrochemical system comprises of a working electrode and a reference electrode which is dipped into a liquid analyte. A voltage signal applied across the electrodes results in a redox reaction at the electrode surface [2]. The presence of multiple compounds in the solution would result in many such redox reactions for each compound at a particular potential called redox potential. Such phenomenon gives rise to a situation where it is very difficult to decide the model order as the system dynamics significantly depends upon the type of compound used as analyte in the experiment. In this paper, an electronic tongue with tea samples as analyte is taken as case study and its model order determination is done by performing system identification on the input-output data points those are recorded in the experiment as explained in Section II. An electronic tongue is an electrochemical instrument used for the quality estimation of liquid samples [3–5]. It comprises of a three-electrode system with an array of working electrodes, a reference electrode and a counter electrode. The electronic tongue in this case works on the principal of pulse voltammetry [6, 7]. A pulsating voltage is applied across the working electrode with respect to the reference electrode. The response current waveform which passes through the counter electrode contains meaningful information about the analyte [8]. These information are treated as the taste of the liquid samples and they are extracted from the waveforms using various features extraction method [8, 9]. This current has contributions of the redox reaction taking place for different compounds at different potential level of the applied voltage pulse. The system identification is used in this paper to estimate the transfer function model with desired combination of pole and zeros so that the model order with maximum model fit can be found out. The system identification is simply finding the model of the system form the system data itself. The data points of the input and the output of system are measured and the model structure is selected. The system identification algorithm is then used to estimate the model parameters. The selection of model structure can be done by two methods. One is the black box approach and the other is grey box approach. In black box approach, various random model structures are tried to find the most suitable one for the system as the information about model structure is completely absent initially. On the other hand, in grey box approach, the model structure is inspired from some mathematical derivation or some pre-existing requirement of a particular model. The work aims at finding a method to determine the model order of the electronic tongue under observation with the help of system identification approach. The decision regarding the optimized model order is taken after identifying different transfer function of different order and their comparison based on the model fit. The results are obtained for all the tea samples and the working electrodes.

Determination of Model Order of an Electrochemical System …

31

2 Hardware and Experimentation The electronic tongue used in the experiment is a voltammetric electronic tongue which has specially been designed for quality estimation of black tea samples and was used earlier in works [9, 10]. The electrochemical cell present in this electronic tongue has an array of 5 working electrodes made up of noble metals. These electrodes are made up of gold, iridium, palladium, platinum and rhodium. These noble metal electrodes were used as they are not easily affected by the environmental conditions. The reference electrode is an Ag/AgCl electrode while the counter electrode is made up of stainless steel. The working electrode diameter is 1 mm and only 2 mm of their length is exposed into the liquid. The electrode assembly arrangement is such that the distance between reference electrode and all working electrodes are same. Thus the estimated model parameters would only depend upon the change in working electrode and the tea sample but not on the distance between electrodes. The electrode assembly is dipped into the tea liquor sample of 20 mL in volume. The steps followed for the preparation of tea sample in laboratory is similar to the method shown in [10]. In [9], a special type of voltage signal called Large Amplitude Pulse Voltammetry (LAPV) has been used to perturb the electrodes. The same LAPV signal is used in this work as input signal and is applied across the working electrode with respect to the reference electrode. The shape of the LAPV waveform is shown in Fig. 1 where the minimum voltage level is −0.9 V and the maximum voltage is +0.9 V with a pulse width of 40 ms. It has a total of 18 pulses of different voltage levels. This gives the multi-component liquid like tea sample a chance to undergo a large number of redox reactions for most of the compounds present in it at different potential levels. The electrodes are coupled with an electronic device called potentiostat. Its role is to keep constant potentials across the working electrode with respect to the reference electrode. Potentiostat also allow the user to select the potential and its parameters with the help of user interface developed in LabVIEW in computer. It also has an electrode switching circuit that switches between working electrodes in real time and ensure that the LAPV signal is applied to only one electrode at a time sequentially in the order of gold, iridium, palladium, platinum and rhodium. The current generated in the electrochemical reaction is recorded in PC with the help of data acquisition 1

Voltage (volt)

Fig. 1 LAPV voltage signal applied across working and reference electrode

0.5 0 -0.5 -1

0

0.5

1.0

Time (second)

1.5

32

S. Kumar and A. Ghosh

(DAQ) card which is also a part of potentiostat. The DAQ sends the current to the PC in the form of data points sampled at 1000 sample per second. One working electrode produces current waveform which has 1480 data points and a total of 7400 data points are recorded from all the working electrodes in one measurement cycle. All these data points together with the input signal data points are used for the purpose of model estimation using system identification.

3 System Identification and Model Order In this section, the system identification technique is used to find the transfer functions with various model orders of the electronic tongue system. The MATLAB’s (ver. 2017a) system identification toolbox is used here for this purpose that utilizes the measurement data obtained in the experiment to estimate the model parameters. A. System Identification As mentioned earlier that the system identification uses the system data to estimate the model of the system so as to arrive at a model that best explains the system dynamics. Also, the model structure is decided by the grey-box approach here. The initial model structure is decided on the basis of Randles model [2] and from there different pole-zero combinations are tried and the corresponding model estimation is done. The next step of system identification is to estimate the parameters of the identified structure. In this work, the Gauss-Newton (GN) estimation algorithm has been used to estimate the model parameters in MATLAB. The GN algorithm mainly minimizes the cost function which is a function of error between the measured response and the estimated response of the model. The cost function here is defined by the normalized root mean square (NRMSE). The 1-NRMSE is the degree of goodness of fit or simply the model fit expressed in percentage which is given by, model fit (%) = {1 - [ n  1 (ymeas − ymodel )2 / n  1 (ymeas − y’meas )2 ]1/2 } × 100% where, ymeas = measured output of system, ymodel = estimated output of model, y’meas = mean of measured data points, n = number of measured data points. The system identification estimation algorithm fine-tunes the model parameters such that the NRMSE is minimized. The model developed as a result of this fits to the measured response to the maximum possible extent for a particular model structure. Let us suppose that the identified transfer function is given by G. Therefore, ymodel (t) = Gu(t). The error is given by e(t) = ymeas (t) − ymodel (t). To determine the transfer function G, the system identification algorithm minimizes the value of e(t).

Determination of Model Order of an Electrochemical System …

33

Table 1 Possible model structures with model orders Number of zeros

Number of poles

Model order

Model structure of G(s)

Number of unknowns

1

1

1

Z1P1

3

1

2

2

Z1P2

4

2

2

2

Z1P2

5

2

3

3

Z2P3

6

3

3

3

Z2P3

7

3

4

4

Z3P4

8

4

4

4

Z3P4

9

4

5

5

Z4P5

10

5

5

5

Z4P5

11

5

6

6

Z5P6

12

6

6

6

Z5P6

13

B. Model Order of Electronic Tongue The model structure of a typical electrochemical system has been given by Randles model. The transfer function model for the same has been derived by decomposing the Randles equivalent circuit [11]. The Randles model has a 1 pole and 1 zero transfer function which is a first order model. This transfer function is for a single redox pair chemical compound present in the analyte. For multiple compounds, those undergo redox phenomenon during chemical reaction, several such Randles model has to be cascaded to model the complete response of electronic tongue. But in practice, the models with higher orders are complex and their analysis is cumbersome. There has to be some limit of increasing the model order of the electrochemical system like electronic tongue even if the analyte is a multiple component solution like black tea liquor. In this work, starting from the 1 pole-1 zero transfer function, several other polezero combinations are identified and the corresponding model fit is observed. All such pole-zero combinations are shown in Table 1 along with the model structure and the unknown parameters. These models are identified for the entire tea sample and for all the working electrodes. The model orders have been considered up to a combination of 6 poles and 6 zeros.

4 Results and Discussions The system identification has been performed in MALTAB ver. 2017a using the input-output data points obtained from three different liquor of black tea samples (S1, S2 and S3). All the model structures shown in Table 1 have been used and the corresponding model fit are estimated for all the tea samples and with all the 5 working electrodes sequentially. The model fit based on NRMSE is shown in Table 2

34

S. Kumar and A. Ghosh

Table 2 NRMSE based percentage model fit for different tea samples Pole/zero/order

Tea sample

Model fit (%) for working electrodes Au

Ir

Pd

Pt

Rh

1 pole, 1 zero order: 1

S1

76.73

71.83

73.41

72.78

54.44

S2

75.46

72.46

76.28

77.36

75.08

S3

75.14

70.75

74.29

73.01

73.41

2 pole, 1 zero order: 2

S1

78.10

85.58

76.44

80.12

81.39

S2

77.80

85.42

78.59

81.61

82.50

S3

77.96

85.44

76.93

79.95

82.34

2 pole, 2 zero order: 2

S1

76.99

91.39

77.04

74.25

76.05

S2

76.11

92.35

78.47

88.96

88.23

S3

76.78

91.65

77.04

74.38

87.18

3 pole, 2 zero order: 3

S1

80.37

85.80

80.35

81.20

86.22

S2

78.26

86.00

80.56

82.11

83.52

S3

78.18

85.80

79.93

80.72

82.64

3 pole, 3 zero order: 3

S1

84.83

92.44

82.29

86.12

92.11

S2

84.35

93.29

85.14

89.73

91.23

S3

87.01

92.04

85.68

85.46

92.11

4 pole, 3 zero order: 4

S1

81.11

86.26

79.73

80.76

84.78

S2

79.05

70.80

76.96

81.34

83.65

S3

80.14

86.72

79.58

79.91

84.59

4 pole, 4 zero order: 4

S1

89.19

89.83

83.58

86.82

92.69

S2

85.78

93.37

85.56

91.52

90.84

S3

87.68

92.09

83.63

86.32

90.41

5 pole, 4 zero order: 5

S1

77.79

70.71

82.48

83.00

86.76

S2

80.6

87.88

27.71

67.37

86.49

S3

−86.70

−43.10

82.38

81.10 −34.60

5 pole, 5 zero order: 5

S1

89.91

82.01

88.56

87.19

90.92

S2

85.98

93.41

90.31

91.84

90.79

S3

88.02

92.12

87.7

86.5

88.56

6 pole, 5 zero order: 6

S1

82.57

−68.2

−45.7

−68.8

65.31

S2

80.92

−63.4

53.05

44.83 −20.4

S3

76.89

−79.2

25.99

60.97

69.41

6 pole, 6 zero order: 6

S1

90.65

86.08

82.75

17.40

90.66

S2

89.50

6.23

16.19

73.52

92.36

S3

85.66

48.96

73.94

75.54

88.93

Mean 72.83

80.68

81.79

82.11

88.26

81.02

88.62

49.99

88.92

14.28

67.89

Determination of Model Order of an Electrochemical System …

35

for all the case of tea sample and working electrode considering all the model orders by varying the number of pole and zeros. From Table 2, it is observed that the percentage model fit depends significantly upon the tea sample which is considered and the working electrodes from which the measurement is taken. The model order also has a lot to do with the electrodeelectrolyte combination. A particular model order with a given pole-zero combination may give good model fit for a tea sample or working electrode. While the same model order many not work for other tea sample and working electrode. For instance, the 2 pole-2 zero combination has produced a very good model fit (greater than 90%) for all the tea samples but only in the case of iridium as working electrode. On the other hand, rest of the working electrodes do not show significant model fit for this polezero combination. Also, for a 3 pole-3 zero combination, the iridium and rhodium electrodes shown comparable model fit for all the tea samples. While, in the 2 pole-2 zero model, the fit for iridium is still comparable to that of 3 pole-3 zero model. However, in case of rhodium electrode, the model fit is quite less for 2 pole-2 zero structure compared to 3 pole-3 zero structure. In the model fit observations for different tea sample shown in Table 2, it can be interpreted that for a multi-compound solution like tea, the model fit gets better with increase in model orders. This is evident from the fact that in a multi- compound solution, multiple redox phenomenons occurs. For all such redox reactions, the Randles transfer function model with 1 pole-1 zero are cascaded that increases the order of overall system. That is why the model fit increases when the chosen model structure is of higher order in the case of tea samples. However, there is always a stage where the increase in model order does not help and the model fit starts to decrease or anonymous results are observed. In Table 2, the model fit drops significantly from 5 pole-4 zero combination with an exception at 5 pole-5 zero combinations. It is observed that the order which is acceptable in case of tea samples is from 2 to 4 based upon the requirement. The 3 pole-3 zero, 4 pole-4 zero, 5 pole-5 zero combinations have similar results and therefore, taking higher system order into consideration is not feasible and will increase model complexity. The 3 pole models may be preferred over other models as the transfer function with small orders is easy to analyze. However, depending upon the application, a particular model order can be selected even if the model fit is not on the higher side. In this case, if a lower model is desired, the 1 pole-2 zero combination seems to be a good candidate as far as the electrochemical model with tea samples are considered. The general trend of model fit can be seen in Fig. 2 in which the mean model fit is shown for all the cases of model structure. As mentioned above, the common trend is the increase in model fit with model order and at some instant the increase in model order produces undesirable model fit. Another things to note down is that the model fit is very good in those cases where the number of poles are equal to number of zeros. In Fig. 2, most of the peaks showing model fit correspond to such situation. This is more prominent for higher model orders.

36 100 90 80

model fit %

Fig. 2 Mean model fit change observed by varying pole-zero combination. Here z1 p1 denotes a 1 zero-1 pole combination and so on

S. Kumar and A. Ghosh

70 60 50 40 30 20 10 0 z1 p1

z1 p2

z2 p2

z2 p3

z3 p3

z3 p4

z4 p4

z4 p5

z5 p5

z5 p6

z6 p6

Pole-zero combination

The response of model for model order 1 to 4 is shown in Figs. 3, 4, 5 and 6, respectively, for the case of gold electrode only for tea sample S3. Response of model order 5 and 6 has not been shown as they seem insignificant in this case. In Fig. 3, the response of the 1 pole-1zero model is compared with the actual measured waveform obtained from tea sample S3 with gold electrode. The model fit comparison among 2nd, 3rd and 4th order model configurations are shown in Figs. 4, 5 and 6, respectively.

6

10

-5

measured 1 pole-1 zero: 75.14%

4 2

Current (µA)

Fig. 3 Comparison of model fit between measured response and response of model with 1 pole-1 zero configuration (gold electrode; Sample S3)

0 -2 -4 -6 -8 0.2

0.4

0.6

0.8

Time (seconds)

1

1.2

1.4

Determination of Model Order of an Electrochemical System … Fig. 4 Model fit comparison among 2nd order model configurations

6

10

37

-5

measured 2 pole-1 zero:77.96% 2 pole-2 zero: 76.78%

4

Current (µA)

2 0 -2 -4 -6 -8

0.2

0.4

0.6

0.8

1

1.2

1.4

1

1.2

1.4

1

1.2

1.4

Time (seconds)

Fig. 5 Model fit comparison among 3rd order model configurations

6

10

-5

measured 3 pole-2 zero: 78.18% 3 pole-3 zero: 87.01%

4

Current (µA)

2 0 -2 -4 -6 -8 0.2

0.4

0.6

0.8

Time (seconds)

Fig. 6 Model fit comparison among 4th order model configurations

6

10

-5

measured (y1) 4 pole-3 zero: 80.14% 4 pole-4 zero: 87.68%

4

Current (µA)

2 0 -2 -4 -6 -8

0.2

0.4

0.6

0.8

Time (seconds)

38

S. Kumar and A. Ghosh

5 Conclusion In this work, the system identification based approach has been used to determine the model order of electronic tongue. The work presents a very simple method to know about order of an electrochemical system. The experimental measurements of the system input and the output were used for system identification in which various model order structures were identified. For the case of tea samples, the model order from 2 to 4 is found suitable for analysis purpose. The method discards the 5th and 6th model order structure as the model fit was missing in these cases. The identified model orders of the electronic tongue with tea samples has an average model fit in the range of 80.68 and 88.62%. The paper also discusses on dependency of electrode and analyte on determination of model order of the electrochemical system. Such approach of model order determination can also be extended in other systems as well in which the input and the corresponding response data points can be accessed.

References 1. Karel J (2011) Keesman, system identification: an introduction. Springer, London 2. Bard AJ, Faulkner LR (2001) Electrochemical methods: fundamentals and applications, 2nd edn. Wiley, New York 3. Saha P, Ghorai S, Tudu B, Bandyopadhyay R, Bhattacharyya N (2014) A novel technique of black tea quality prediction using electronic tongue signals. IEEE Trans Instrument Meas 63(10):2472–2479 4. González-Calabuig A, del Valle M (2008) Voltammetric electronic tongue to identify Brett character in wines. on-site quantification of its ethylphenol metabolites. Talanta 179(11):70–74 5. Bhondekar AP, Vig R, Gulati A, Singla ML, Kapur P (2011) Performance evaluation of a novel iTongue for Indian black tea discrimination. IEEE Sens J 11(12):3462–3468 6. Ivarsson P, Holmin S, Hojer NE, Krantz-Rulcker C, Winquist F (2001) Discrimination of tea by means of a voltammetric electronic tongue and different applied waveforms. Sens Actuators B Chem 76(1–3):449–454 7. Ivarsson P, Krantz-Rülcker C, Winquist F, Lundström I (2005) A voltammetric electronic tongue. Chem Sens 30(1):i258–i259 8. Palit M, Tudu B, Dutta PK, Dutta A, Jana A, Roy JK, Bhattacharyya N, Bandyopadhyay R, Chatterjee A (2010) Classification of black tea taste and correlation with tea taster’s mark using voltammetric electronic tongue. IEEE Trans Instrum Meas 59(8):2230–2239 9. Ghosh A, Sharma P, Tudu B, Sabhapondit S, Baruah BD, Tamuly P, Bhattacharyya N, Bandyopadhyay R (2015) Detection of optimum fermentation time of black ctc tea using a voltammetric electronic tongue. IEEE Trans Instrum Meas 64(10):2720–2729 10. Ghosh A, Bag AK, Sharma P, Tudu B, Sabhapondit S, Baruah BD, Tamuly P, Bhattacharyya N, Bandyopadhyay R (2015) Monitoring the fermentation process and detection of optimum fermentation time of black tea using an electronic tongue. IEEE Sens J 15(11):6255–6262 11. Kumar S, Ghosh A, Tudu B, Bandyopadhyay R (2017) An equivalent electrical network of an electronic tongue: a case study with tea samples. In: 2017 ISOCS/IEEE international symposium on olfaction and electronic nose (ISOEN), Montreal, QC, pp 1–3

Signal Processing

Problem Diagnostic Method for IEC61850 MMS Communication Network Anjali Gautam and S. Ashok

Abstract This paper describes the analysis of the IEC61850 MMS and GOOSE communication network using a laboratory setup. The MMS communication between the OPC server and simulated IED is established and communication is captured using the open-source tool Wireshark. The normal flow of communication is analyzed and decoded first and IEC61850 data is manipulated in IED to determine how the communication flow deviates from the standard flow of communication. Reporting to the Station HMI and SCADA is done using MMS communication service on the Ethernet network. To determine the status and quality of the IEC61850 data exchanged between IED’s (GOOSE), IED and HMI (MMS), Wireshark is used to capture the network traffic between these two scenarios. Moreover, these captured scenarios in Wireshark are used to diagnose whether the error is a configuration error or the network error. These Wireshark log files are sent by the users of the IED’s in the substation to the vendors of IED to diagnose the error codes. The efforts in diagnosing the errors can be reduced if one knows the flow in the normal scenario and abnormal scenarios which can help to reduce the time to troubleshoot the IEC61850 communication network. Keywords MMS (Manufacturing Message Service) · GOOSE · Wireshark · OPC server · IED

1 Introduction IEC61850 [1] is a widely known standard used in substation automation system all around the world nowadays. It is not limited to the communication interface in substation but also how electrical devices are modeled (in the form of data based on the OOPS concept) in one standard form and mapping the information models on legacy A. Gautam (B) · S. Ashok Department of Electrical Engineering, NIT Calicut, Kozhikode, India e-mail: [email protected] S. Ashok e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_4

41

42

A. Gautam and S. Ashok

protocols using the independent services called as abstract services mentioned in the standard. There were many protocols available for mapping information model [2]. This protocol comes into the picture to satisfy the need for interoperability between different vendor IEDs. It is the outcome of the efforts of UCA and IEC community to come up with one international standard in which any vendor devices can work independently with other vendor devices with the flavor of emerging information technology in communication fields [3].

2 Outline of IEC61850 A. Evolution The process of development of substation automation protocol involves efforts and foundation work of many stages that involves UCA and IEC technical committee. It was a long process as the work initiated by the UCA group became the stepping stone and then a set of recommended protocols for the ISO communication layer models were suggested. Thereafter the architecture is developed under UCA standard that consists of a data object model and services. Finally, IEC technical committee working groups 10 results in the final outcome that is known today as IEC61850 [4]. B. Parts of IEC61850 protocol The standard is subdivided into 10 parts which are the nutshell of substation communication aspects. Part 1 and 2 consist of an introduction and glossary for the standard. Part 3, 4, and 5 provide the requirement of substation communication and system management. Part 6 provides the XML-based configuration language used in substation also called SCL (Substation Configuration Language). Part 7 provides the different aspects like principle and data model and services and logical nodes and data classes of the feeder and its equipment in the substation. Part 8 describes the mapping of data objects and ASCI services to MMS. Part 9.1 defines the mapping of sampled values from the actuators, sensors, and status value to the protocol multidrop point to point link (unidirectional) and part 9.2 specifies the mapping of sampled value over ISO/IEC 88203 with user configured dataset. Part 10 specifies the conformance testing. C. Communication in substation Standard IEC61850 provides three communication services; these are connectionoriented MMS, connectionless GOOSE, and SMV. According to the architecture of the substation, these services are also called vertical and horizontal, respectively.

Problem Diagnostic Method for IEC61850 …

43

(1) MMS IEC61850 data and services can be mapped to any protocol but MMS is chosen among other protocols because its makes mapping less complex and cumbersome and it is able to handle complex structure of object naming and services in the IEC61850 [5]. It maps ASCI services like server into service defined in MMS called domain and so on. MMS communication handles medium priority data like reporting and logging between IED and OPC clients like SCADA and HMI. (2) GOOSE This information consists of the interlocking signal, trip signals, status information, and warning signals [6]. The signals mapped on the data link layer and operating capacity is faster than MMS as it maps to all seven layers of the ISO communication model. D. Substation architecture based on IEC61850 The architecture of the substation is divided into three levels called process, bay, and station from field-side view. It also consists of station and process Ethernet LAN bus which comprises information/data from the process and bay level, respectively [7]. The data from the instrument transformer, sensors, and actuators are transmitted to the merging unit to be digitized and it is available on the process bus using redundant optical-fiber Ethernet LAN. The process bus data is ready to transfer on IEDs present in a bay level that can further be transferred to the station bus using MMS or GOOSE communication. The data from the station bus can be sent to the OPC server and that can be further imported to the OPC clients like HMI and SCADA and to the remote location of another substation architecture shown in the Fig. 1. E. Modeling Approach Modeling approach utilized in IEC61850 standard not only defines how data is transferred on the wire but also describes how protection device data should be organized [8] independent of the vendor. This approach is utilized to achieve the goal of interoperability between different vendor devices and also it saves the configuration efforts. For example, if one put the value of CT/PT inside IED then it will automatically assign that value in its associated model so this enables the self-configuration feature of IED. Import of SCL files in IED is also used for configuration of IED which further reduces the configuration effort and also reduces the errors made during configuration between different IEDs. The data modeling of the protection devices and their information involves a sequential collection of logical device, logical nodes, data object, and its attribute as shown in Fig. 2.

44

A. Gautam and S. Ashok

Fig. 1 Substation architecture

The concept of the object model can be explained with the example of circuit breaker logical node which is represented as XCBR in the standard. XCBR contains data that is pos for the position; the data inside the logical node is classified according to the common data class CDC which defines the type and structure of the data. Data contains attributes like stval (status value), q (quality), etc; that are subjected to another group of categorization called functional constraints (FC) like ‘St’ for status.

Problem Diagnostic Method for IEC61850 …

45

Fig. 2 Object model of IEC61850

3 Packet Analysis and Its Need Stability is one of the important features of communication flow in the substation automation system. For this reason, there is the need to analyze the packets present in the network traffic of substation while establishing and operating internal communication between server and client in the substation automation system. Also, this analysis of data is helpful when the measurability of the apparatus and system is required. Substation communication networks carry a large amount of real-time data transfer in terms of MMS, GOOSE, and SMV modes of communication which gives the need to optimize the analysis of the network packet to reduce the time taken in troubleshooting and to speed up runtime diagnostics [9]. Hence the method is proposed in the later section.

46

A. Gautam and S. Ashok

(a) Limitation of the previous methods of analysis For information report on station bus, IEDs sends data to the OPC server that interfaces with the OPC Clients. These data are analyzed by the network capturing tools such as Wireshark, Ethereal, etc. These tools capture all the data transferred in the wire where it is installed and configured to monitor data. These tools capture network traffic and arrange them one by one with the details of the protocols and services used in layers of communication. Extra analysis efforts are needed which not only decompose and arrange packets sequentially but also analyze the flow to determine the errors and causes of the errors like network and configuration error to maintain reliability in a communication network in the substation [10].

4 Troubleshooting Method for IEC61850 MMS Communication In this section, the method is proposed for troubleshooting MMS communication. First, the description of packet analysis and its need is explained. Second, the steps required to establish the communication and configuration setting and then approach on how to analyze and observations made in the analysis of some test cases are done. Finally, the significance and application of the work are done. The block diagram of the method is shown below. Here the output of the process 1 is SCD file which is loaded in process 2 and 3. Established communication between process 2 and 3 is analyzed in process 4 and analysis is the outcome of process 4 which is the base for the case study done to vary the process 1, 2, and 3 independently and checking how it deviates from the standard result which will help to understand the behavior communication in substation automation system (Fig. 3).

Fig. 3 Block diagram

Problem Diagnostic Method for IEC61850 …

47

A. Process 1 The input to the process 1 is an ICD template file given to the system configurator tool (IET 600). Steps involved in the configuration of the ICD file using IET600 tools are mentioned below. • Create a new profile. • Create substation, voltage level, bay level, and IEDs in the explained order. • Create subnetwork bus and map IEDs and configure IP address such that the communication between the two IEDs can be set up. • Create a dataset for reporting and map it to RCB block. Configure RCB client with the OPC server IED. • Check for consistency and export the SCD file. B. Process 2 Process 2 involves steps to configure OPC server those are explained below. • • • • •

Create project-create computer node-create IEC61850 OPC server. Import the SCD file. Change the report control identity same as used in process 1. Change the IP address of the network adapter same as mentioned in the process 1. Upload, reload, and check whether init file is created after updating and reload.

C. Process 3 Process 3 consists of simulated IED which can be configured in the following way. • • • •

Create a new profile. Browse process 1 SCD file and set the same IP address as that of process 1. Launch simulated IED. In process 2, open diagnostic AE log file to confirm the status of connection establishment.

D. Process 4 Process 4 involves the analysis of communication in Wireshark software. Steps involved are mentioned below. • Start Wireshark and select the interfacing adapter. • Apply the capture filter to avoid unnecessary network traffic. Packets captured are shown in the below Fig. 5. • Analyze packets step by step. (a) Analysis of Packets Captured • ARP request and response for identifying the interfacing process through their MAC address and TCP three-way handshaking for establishing communication in initial packets, here in packets 1–7.

48

A. Gautam and S. Ashok

• COTP connection request for transfer data reliably on the top of TCP to add the same set of boundaries as send by the sender in packets 8–9. • Initializing MMS communication by sending initiate request and response in packets 10–11. • MMS confirmed request and response using getNameList service to get the logical device present in the IED communicating in packets 13–14. • Reading RCB (report controlling block) structure using “getVariableAcessAttribute” service. Here, we can identify whether the report is buffered or unbuffered and also the response number of the control block attributes which are included in the SCD files can be identified from the response PDU in the packets 16–17. • Reading data of RCB like RptId, RptEna, DatSet, ConfRev, OptFlds, BufTm, SqNum, IntgPd, GI, PurgeBuf, EntryID, and TimeofEntry attributes in packets 16–17 using service get-variable-access- attributes. • Reading values of the RCB data using service get- variable-access-attribute. Here the data types mentioned in IEC61850 is mapped to data types like octet string, bit string, etc. in packets 19–20. • Reading structure of the dataset in packets 22–23. • Reading attributes of the dataset structure item in packets 25–26. • Configuring RCB attributes for reporting using write service in packets 28–29. • MMS write for creating Entry Id in packets 31–32. • Enabling report using MMS write in packets 34–35. • MMS unconfirmed PDU—this report is sent to the OPC server from the IED periodically or depending upon the triggering condition. Analysis of the unconfirmed PDU is shown below. Figure 4 shows the analysis of the information report. It contains information in terms of the access results. The number of access results depends upon the dataset entries and buffered or unbuffered report. Access result 1 represents the data type visible string which gives information about the RptId. Access result 2 represents the data type bit string which gives the information about Opt Fld. Access result 3 represents the data type unsigned which gives information about the sequence number. Access result 4 represents a time of entry (binary time). Access result 5 represents the data type Boolean which gives the information about BufOvflw (buffer overflow) (Fig. 5). Access result 6 represents data type octet string and gives information about Entry Id. If the entry is just made after communication establishment, then it starts from 0. Access result 7 represents the data type bit string which gives information about inclusion bit string. From Access result in 8 onwards, the information about the dataset and its entries are shown. If the dataset contains two entries, then Access result 9 will come, order, and go on increasing. Here in this file, the dataset consists of the stval, quality, and time attributes.

Problem Diagnostic Method for IEC61850 …

Fig. 4 Information report

49

50

Fig. 5 Packet captured

A. Gautam and S. Ashok

Problem Diagnostic Method for IEC61850 … Table 1 Representation of quality bit in Wireshark

51

Quality

Simulator (type-bit string)

|CET tool

Good

0000

0

Invalid

4000

1

Invalid, overflow

6000

1

Invalid, bad reference

4800

1

Invalid, oscillatory

4400

1

Invalid, failure

4200

1

Invalid, old data

4100

1

Invalid, inconsistent

4080

1

Invalid, inaccurate

4040

1

Invalid, out of range

5000

1

Invalid, all checked in set detail

7fc0

1

Reserved

8000

2

Reserved, overflow

A000

2

Reserved, bad reference

8800

2

Reserved, oscillatory

8400

2

Reserved, failure

8200

2

Reserved, old data

8100

2

Reserved, inconsistent

8080

2

Reserved, inaccurate

8040

2

Reserved, out of range

9000

2

Questionable

C000

3

Questionable, overflow

E000

3

Questionable, bad reference

C800

3

Questionable, oscillatory

C400

3

Questionable, failure

C200

3

Questionable, inaccurate

C040

3

Questionable, out of range

D000

3

E. Case study Case 1: Make a change in simulated IED in process 3 by changing a quality bit. In Wireshark, for different quality, the values will be different as shown in Table 1 and which bit is affecting due to the change in quality that diagnosis is given in Table 2.

52

A. Gautam and S. Ashok

Table 2 Pattern observed for quality bit Good

0000

0000

0000

0000

Invalid

xlxx

xxxx

xxxx

xxxx

Reserved

lxxx

xxxx

xxxx

xxxx|

Questionable

llxx

xxxx

xxxx

xxxx

Overflow

xxlx

xxxx

xxxx

xxxx

Out of range

xxxl

xxxx

xxxx

xxxx

Bad reference

xxxx

lxxx

xxxx

xxxx

Oscillatory

xxxx

xlxx

xxxx

xxxx

Failure

xxxx

xxlx

xxxx

xxxx

Old data

xxxx

xxxl

xxxx

xxxx

Inconsistent

xxxx

xxxx

lxxx

xxxx

Inaccurate

xxxx

xxxx

xlxx

xxxx

Test

xxxx

xxxx

xxxl

xxxx

Operator blocked

xxxx

xxxx

xxxx

lxxx

Process

0000

0000

0000

0000

Substituted

xxxx

xxxx

xxlx

xxxx

Case 2: Time stamp mismatch in OPC server and simulator. As we changed simulator machine date to 1 May (past time), Access result in UPDU 4 will show the changed simulator machine time and Access result 8 will show the time of the simulator when we loaded the SCD file. i.e., May 16, 2018, 07:15:073000013 as shown in Fig. 6. Case 3: No change in IEC61850 data in simulator. When no change occurred in the IEC61850 data present in simulated IED, IED kept on sending a signal to maintain the communication between the OPC server and IED. This can be verified in the Wireshark tool by observing the continuous sending of TCP Keepalive from IED to the OPC server. Case4: For SCD without a dataset. For this case, the information report is not sent and the communication is concluded after reading the logical device information. And only less exchange of services is observed that can be compared with the result of the process 4 as shown in Fig. 7.

Problem Diagnostic Method for IEC61850 …

Fig. 6 Test case 2 Wireshark screenshot

Fig. 7 Case 4 packet captured

53

54

A. Gautam and S. Ashok

5 Conclusion IEC61850 is the backbone of the substation automation system. It is needed to understand the behavior of the IEDs for normal and abnormal data produce by either manipulation in the input of the IED or by changing different types and values like quality, etc. This paper describes how we can perform these tests in detail from the beginning initial configuration in a file in different software to the verification in the Wireshark tool. Reporting service from the IED to the OPC server that interface with the HMI is analyzed in detail. This information is sent for monitoring the status of the IED in HMI or remote location.

Reference 1. Yu MJ, Jung JH, Choi HS, Lee JS (2015) Implementation and performance measurement of a packet analyzer for traffic monitoring in tactical communication network. In: Proceedings of the Korea Institute of Military Science and Technology, Korea, June 2015, pp 1015–1016 2. Netted Automation (2002) The MMS client/server model [cited 10 Aug. 2015]. http://www. nettedautomation.com/standardization/iso/tc184/sc5/wg2/mms_intro/intro3.html> 3. IEC 61850-1 (2013) Communication networks and systems for power utility automation – introduction and overview. IEC Technical report, Edition 2.0, 2013-03 4. IEC 61850—Communication networks and systems in substations. http://domino.iec.ch/ webstore/webstore.nsf/searchview/?SearchView=&SearchOrder=4&SearchWV=TRUE& SearchMax=1000&Query=61850&submit=OK 5. IEC 61850-8-1 (2011) Communication networks and systems for power utility automation – specific communication service mapping (SCSM) – Mappings to MMS (ISO 9506-1 and ISO 9506-2) and to ISO/IEC 88023, IEC International Standard, Edition 2.0, 2011-06 6. Kriger C, Behardien S, Retonda-Modiya J (2013) A detailed analysis of the GOOSE message structure in an IEC 61580 standard-based substation automation system. Int J Comput Commun Control 8(5):708–721. ISSN 1841-9836 7. Adamiak M, Baigent D, Mackiewicz R (2010) IEC 61850 communication networks and systems in substations 8. Kunz G, Machado J, Member, IEEE, Perondi E, Vyatkin V (2017) A formal methodology for accomplishing IEC 61850 real-time communication requirements. IEEE Trans Ind Electron 64(8):6582–6590 9. Wimmer W, Rietheim (2010) United State Patent No. US 2010/00399.54 Alexzendria. ABB Technology, Zurich 10. Premaratne UK, Samarabandu J, Sidhu TS, Beresh R, Tan J-C (2010) An intrusion detection system for IEC61850 automated substations. IEEE Trans Power Delivery, 25(4):2376–2383

IntelliNet: An Intelligence Delivery Network Dipnarayan Das and Sumit Gupta

Abstract As per researchers, “AI research” is defined as the study of intelligent agents, where an intelligent agent is any device that perceives its environment and takes actions that maximize its chances of successfully achieving its goals. Through this paper, we are proposing IntelliNet, which is an Intelligence Delivery Network (IDN). The objective of such a network is to infect intelligence into all the objects or devices which will be connected with the network. The prime reason behind infecting intelligence and not injecting it into any object is that here the source intelligence system not only delivers intelligence to the object but also senses all types of possible events for gathering knowledge in order to infect other objects as well, thereby making ordinary devices smart and intelligent. Keywords Artificial Intelligence · Machine Learning · Natural Language Processing · Knowledge Base · NodeMCU

1 Introduction Artificial Intelligence (AI), sometimes called Machine Intelligence, is intelligence demonstrated by machines, which is in contrast to natural intelligence displayed by humans and other animals [1]. Many so-called smart devices exist in our lives but the smartness or intelligence shown by them is static and database-oriented. These devices, in reality, do not possess or show quick-witted intelligence but just work on the basic information-retrieval or content-retrieval methodology. Our system attempts to add real intelligence and smartness to the devices. We are attempting to convert normal devices into their intelligent counterparts when the devices come in contact with the intelligent environment. To do so, we D. Das (B) · S. Gupta Department of Computer Science & Engineering, University Institute of Technology, The University of Burdwan, Golapbag (North), Burdwan 713104, West Bengal, India e-mail: [email protected] S. Gupta e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_5

55

56

D. Das and S. Gupta

have used the Cybernetics approach in our system to make it work like GOFAI (Good Old-Fashioned Artificial Intelligence), a term coined in 1985 by John Haugeland [2]. Cybernetics is a trans-disciplinary approach for exploring regulatory systems, their structures, constraints, and possibilities. Norbert Wiener defined Cybernetics in 1948 as “the scientific study of control and communication in the animal and the machine” [3]. We have used the notion of Deep Learning, which is a well-known paradigm of Artificial Neural Network based Machine Learning. A Deep Learning system is capable of learning a long chain of causal links by itself. Our intelligent system uses the concept of Deep Learning so that it can learn about relational and eventual incidents very quickly and easily. Here, Deep Learning has been used to transform Natural Language Processing (NLP) into transparent, human-like AI. This system is built keeping in mind the linguistics. Linguistics is the scientific study of language and involves an analysis of language form, language meaning, and language in the context of the system so that it can possess a self-adapting ability to grow in the real world [4]. In this paper, we have presented our intelligent system whereby we have tested the activities of our proposed Intelligence Delivery Network (IDN) on the self-made normal watch. What we have observed is that when the normal watch is connected to the network, it gets converted into a smart watch possessing novel features like Speech Synthesis, Text to Speech Conversion, Live streaming, and Web Surfing. The rest of the paper is organized as follows: Sect. 2 presents our proposed work in the form of system architecture, system components, theories related to the system implementation, an algorithm showing the different steps followed to achieve the objective, and working principle of our system along with different aspects of its implementation. Section 3 deals with the discussion and analysis where we have focussed on the key points of our system with special emphasis on where the present systems lack and how our proposed system deals with the challenges posed by current systems. Section 4 concludes the paper followed by references in the end.

2 Our Proposed Methodology A. System Architecture: The architecture of our proposed Intelligence Delivery Network (IDN) named IntelliNet is depicted in Fig. 1. Here the intelligence is made in a cross platform to share knowledge among all devices connected in the network. The architecture of our proposed smart watch is depicted in Fig. 2. Here, the intelligence is delivered through the Network Unit. B. System Components: To design and implement our Intelligent Delivery Network, we have used various electronic components which are discussed as follows:

IntelliNet: An Intelligence Delivery Network

Fig. 1 Our proposed system architecture

Fig. 2 Our proposed smart watch architecture

57

58

D. Das and S. Gupta

1. Node Micro Controller Unit: NodeMCU (see Fig. 3) is an eLua based firmware for the ESP8266 WiFi SOC from Espressif. The NODEMCU firmware is a companion project to the popular NodeMCU dev kits, ready-made open source development boards with ESP8266-12E chips [5]. 2. SD card Module: The Arduino SD Card Shield (see Fig. 4) is a simple solution for transferring data to and from a standard SD card. The pinout is directly compatible with Arduino, but can also be used with other microcontrollers [6]. 3. Nokia 5110 display Module: Nokia 5110 display module (see Fig. 5) consists of a Serial Peripheral Interface and the display is monochromatic. 4. Heart Rate Sensor: The heartbeat sensor (see Fig. 6) is based on the principle of photo phlethysmography. It measures the change in volume of blood through any organ of the body which causes a change in the light intensity through that organ (a vascular region). In the case of applications where the heart pulse rate is to be monitored, the timing of the pulses is more important. The flow of blood (volume) is decided by the rate of heart pulses and since light is absorbed by blood, the signal pulses are equivalent to the heart beat pulses [7]. C. Related Theories: IntelliNet, our proposed Intelligence Delivery Network is a fully autonomous and self-adapting system that is capable of imparting and delivering intelligence to all Fig. 3 NodeMCU

Fig. 4 SD card adapter

IntelliNet: An Intelligence Delivery Network

59

Fig. 5 NOKIA 5110

Fig. 6 Heart rate sensor

the devices connected to it. This system acts as a creator for its self-made artificially intelligent devices. As it is known in the real world that attributes like intelligence and common sense are imparted in each of us by our creator; similarly, our proposed system IntelliNet implements the attributes of common sense and intelligence to its peripheral or neighboring devices. In fact, our system follows the same approach which a normal being adheres to; mainly the activities that involve different stages of learning, understanding, and analyzing situations from the environment. It is an open secret that devices like smart watches are actually neither fully smart nor intelligent because they are programmed to perform only some specific tasks. Our proposed IDN creates a force field, where any object that enters its circumference or network horizon has to obey the property specified by the IDN, i.e., IDN can make adapted duplicates into various types of intelligent objects that come in contact with it. Thus, IDN acts as a core system, thereby implying that we need not apply the notion and techniques of Artificial Intelligence each time in every single device for

60

D. Das and S. Gupta

upgrading a normal device into its smart (artificially intelligent) counterpart. On top of it, our proposed system is capable of enhancing its knowledge pool rapidly, thus making it an ideal candidate in the field of defense. As our system can perform NLP tasks, it is self-adaptive, thereby covering most of the different artificial state devices. Also, the IDN will lessen the cost factor by replacing the microcontroller area and the storage area, usually needed in each smart device. The heart rate monitor is used as a component of the smart watch which is the implementation of the device in which intelligence has been infected by the IDN. The IDN can collect various event-related information from the infected devices such as how many devices are in use during a specific year, how many devices have good health statuses, and so on. Using this, we can get an exhaustive statistical report for future purpose while performing any specific work. The smart watch has no intelligence without the IDN. For any speech recognition activity, the normal watch sends the query to the IDN, which will further send an action response in the realtime mode. To understand the query which is in natural language, the Text Mining techniques are used. This upgrades the normal watch into a smart watch. To perform the activities, our system works primarily on the following four types of technological domains, viz., Artificial Intelligence, Machine Learning, Data Mining, and Natural Language Processing. The system contains all the necessary grammars and their rules which are used in our daily life. The grammar which is fed into the system is as follows: 1. Tense a. Present b. Past c. Future 2. Vowels and Consonants 3. Phrasal verbs 4. Articles a. A b. An c. The 5. Parts of speech a. b. c. d. e. f. g. h.

Noun Pronoun Verb Adverb Adjective Preposition Conjunction Interjection

IntelliNet: An Intelligence Delivery Network

61

During the time of initiation or startup, IntelliNet loads all the grammar from its database (which is flat in nature). After that, it initializes the common structural forms of every sentence which will be used during the time of sentence analysis. Our system is made with popular web development technologies. But the major portion of it can be implemented into a Microcontroller environment as in Arduino, NodeMCU, etc. The web development technologies used here are HTML, CSS, JAVASCRIPT, PHP, BATCH, VBSCRIPT, and JQUERY. Moreover, we have not used any of the popular and mostly used AI languages like Prolog, Lisp, MXNet, Haskell, etc. D. Proposed Algorithm: IntelliNet, our proposed Intelligence Delivery Network (IDN) loads sequentially all the different types of grammar from its Knowledge Base (KB) like the static sentences, sentence formats, priority of parts of speech, parts of speech, word relations, word regions (what type of output or decision is generated), etc. The proposed algorithm is as follows: START CREATE GLOBAL VARIABLES FETCH THE CONTENTS FROM THE DATABASE SPLIT SENTENCES INTO TOKENS CONTINUE PARSING UNTIL TOKENIZATION IS COMPLETE CREATE DIFFERENT MATRICES BASED ON GRAMMAR AND THEIR RULES STORED IN KB Step 7: FORM NECESSARY RELATIONS BETWEEN THE MATRICES USING KB AND NLP TECHNIQUES Step 8: GENERATE RESPONSE AND UPDATE KB Step 9: END Step Step Step Step Step Step

1: 2: 3: 4: 5: 6:

E. Working Principle: IDN’s Knowledge Base (KB) is a flat-file model whose metadata is defined as per our proposed algorithm. As already specified, the different kinds of data that exist are parts of speech, articles, phrasal verbs, etc. Since the database follows a flatfile model for defining the relationships, we need to just modify some sequence of parameters that will automatically change the whole relationship chart at all levels. Let us understand this concept using the following example that defines simple relations: i. ii. iii. iv. v.

son = father/mother-child-male brother = father/mother-child-male father = grandfather/grandfather-child-male mother = maternal grandfather/maternal grandfather-child-female sister = brother-female

62

D. Das and S. Gupta

Along with the flat file model, we are also using the Structured Query Language (SQL) structure for processing the query. This depends on the complexity of the information, e.g., if we have to store any relationship information, then we will use the flat file model but if the data is related to any event, then we will use SQL to store all the content words from a sentence. This describes the creation of the Relation schema and the Event schema as shown in Fig. 1. After the process of loading data into the database is complete, our system checks or refreshes the input and the output interfaces until it faces death or is terminated. IDN has its own encoding system in MT Extra font which is likely to be an unsupported font. So at the time of interfacing with our system, the device uses the IDN API to convert its request in the corresponding IDN language. In our tested system, there is only one past memory buffer to fetch data from the memory. For instance, let us explain the entire process using the following example: I am Mohit. Neha is my sibling. Who is Neha?

The IDN will at first scan the first sentence and store the user as “Mohit”. Now for the second sentence, the IDN will process to identify the meaning of the word “sibling”. It will then infer that sibling denotes two persons having the same level of relationship in the same tree relation. So, here in the past memory buffer, IDN stores that “Mohit-Neha is sibling.” Next, when the third sentence “Who is Neha?” is encountered, our system checks the relation of “Neha” with the user “Mohit”. The IDN finds that the past memory buffer has this result stored previously. But in the case when the memory buffer does not have the result, then the IDN will check “Neha’s” relation with any other user who is known to the user “Mohit” and will provide the response through the output interface. As per the given figure (see Fig. 7), we observe that the IDN checks for the sentence type by parsing it at the time of feeding queries. As in the previous example, we can say that the sentence had interrogation as well as simple assertive sentence structures.

Fig. 7 Query interface and identification

IntelliNet: An Intelligence Delivery Network

63

The interrogative sentences will be processed by the IDN depending upon the output region type of the “wh” word present in the sentence. If the sentences are assertive, then the IDN will process the sentence by processing each word which will learn, frame and/or rectify (based on the level of matching accuracy) the response to the specific type which is needed to be found (see Fig. 8). For instance, if the input sentence is “I am very happy,” the IDN will format the above sentence as follows: I am very Pronoun AuxiliaryVerb Adverb

happy Adjective

The vector is made by the auxiliary verb “am” which refers to the pronoun “I”. Here the rectified word which forms the crux of the sentence is the adjective “happy”. So this is a very basic example of processing input sentences using the notion of Natural Language Processing (NLP). At any moment if any data is not available with the IDN, then the IDN will search about the topic in the existing search engines. After searching, the IDN will get a huge amount of data in the form of source codes which will be further processed by using an advanced form of Dynamic Crawling and Machine Learning. The type of crawling process can be changed by updating the configuration file. After processing the facts post the crawling phase, although the generated information will be known to us, it would not be in the machine understandable form. To allow our system

Fig. 8 Resultant region selection

64

D. Das and S. Gupta

Fig. 9 Response generation and knowledge update

(IDN) to interpret the collected information successfully, the information is further processed using the contents of the Knowledge Base (KB) [8] and Natural Language Processing techniques and is further stored into the flat database model by using the mechanisms of Data Mining (see Fig. 9). F. Implementation: Our application is centralized. The main core intelligent component is situated in a centralized web server. To implement our system, we have used Deepnl, which is a neural network Python library especially created for performing tasks of Natural Language Processing. It provides tools for Part-of-Speech (POS) tagging, namedentity recognition, semantic role labeling using Convolution Neural Networks, and word embedding creation. As we have tested the working of our proposed system on a manually created database, there is still an ample scope of checking the system’s performance using benchmark datasets and standard performance measures. Measures such as efficiency, accuracy, error rate, etc., will dynamically change depending upon the size of the Knowledge Base. If the size of the Knowledge Base is increased, then all the dependency relationships will be completely defined and there will be an almost negligible chance of any error or inaccurate output based on unknown data. During that situation, the structure of natural language will be successfully processed using the feature extraction method of Machine Learning.

IntelliNet: An Intelligence Delivery Network

65

3 Discussion and Analysis Based on what we have observed, studied, and analyzed during the research activity revolving around such a multi-disciplinary field where different domains such as Artificial Intelligence (AI), Machine Learning (ML), Data Mining (DM), and Natural Language Processing (NLP) are amalgamated together to create an Intelligence Delivery Network named as IntelliNet, we can list out a few important revelations: 1. The main problem of today’s Artificial Intelligence is that it is mostly static in nature where only keywords and sentences are matched to get the result. Thus, these static devices aim at exploring the field of Information or Content Retrieval alone without stressing enough on the aspect of generating and propagating intelligence. 2. The most important requirement for building an intelligent device or network is that we need to create a huge Knowledge base (KB) that can store the keywords and sentences so that NLP tasks can be performed with accuracy and precision. 3. We can observe in our real life that all the smart devices which we use are basically developed using a prebuilt and/or static database. In fact, the component of Machine Learning which these devices promise to incorporate is in reality driven by just the data and information fed into the database linked with these devices. The striking features supported by our proposed system over the existing systems include 1. IDN’s capability of adapting and collection or gathering rate for different types of data is very high because the network contains multiple nodes. 2. Using only one system, i.e., IDN, we can improve the intelligence of all its nodes or different devices connected to it without the need of employing separate storage components or area, thereby reducing the storage requirement. Thus, our proposed system named IntelliNet is an Intelligence Delivery Network that relies not only on static databases, instead creates a Knowledge Base and utilizes the concepts and methodologies of AI, ML, DM, and NLP for achieving dynamic behavior and showcasing human sort of intelligence.

4 Conclusion This paper presents our proposed system named IntelliNet, which is an Intelligence Delivery Network (IDN). Using IDN, we will be capable of infecting intelligence, i.e., spreading, propagating, and inducing intelligence into the devices that are connected to the network. So, intelligence will be drastically expanded into any device that comes in contact with our IDN environment. It is further observed that the intelligent environment is dynamically updated by updating its Knowledge Base so that our proposed system can help us achieve the real concept of Artificial Intelligence and Machine Learning.

66

D. Das and S. Gupta

References 1. Poole D, Mackworth A, Goebel R (1998) Computational intelligence: a logical approach. Oxford University Press Inc., New York 2. Haugeland J (1985) Artificial intelligence: the very idea. MIT Press, Cambridge, MA 3. Wiener N (1948) Cybernetics: or control and communication in the animal and the machine. MIT Press, Cambridge, MA 4. Halliday MAK (2006) On language and linguistics. Continuum International Publishing Group, New York, USA 5. Benchoff B (2015) A dev board for the ESP Lua interpreter. https://hackaday.com/2015/01/01/ a-dev-board-for-the-esp-lua-interpreter/. Accessed 1 Jan 2015 6. Arduino webpage https://www.arduino.cc/en/Reference/SD. Accessed 25 July 2018 7. Elprocus webpage https://www.elprocus.com/heartbeat-sensor-working-application/. Accessed 25 July 2018 8. Hayes-Roth F, Waterman DA, Lenat DB (1983) Building expert systems. Addison-Wesley, Reading

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System Sumit Gupta and Puja Halder

Abstract Text mining is the process of extracting and/or deriving high-quality information from unstructured text by properly structuring the raw text. The structured texts serve as ideal candidates for revealing the syntactic and semantic interpretations encapsulated in them. To do so, we need to employ different methods of text mining and text analytics. Text Analytics deals with the objective of evaluating and assessing text by the application of natural language processing and other linguistic-oriented analytical methods. Text-based sentiment analysis aims to determine the attitude and sentimental state of an author by analysing different tokens of the texts in terms of their polarity. The aim of this paper is to propose a hybrid Lexicon-based sentiment and behaviour prediction system which can help one to comprehend the sentimental as well as the behavioural context of the author. We have used two sets of lexicons, viz. SenticNet 4.0 Lexicon and our own manually created Profile Lexicon in order to assess the input text and to predict the sentiment conveyed by the text as well as to identify the behaviour of the author. Our system works fairly in case of predicting both sentiment and behaviour by offering an accuracy of approximately 90%. Such a system has immense potential in identifying the real intention of an author once the behavioural and sentimental patterns of an author are predicted consummately. Keywords Text mining · Text analytics · Behaviour mining · SenticNet · Profile Lexicon

1 Introduction One of the most significant applications of data mining is to mine the characteristics of the writer or author, which is hidden under the complex textual structure. The procedure to predict such a behavioural pattern is termed as behaviour mining S. Gupta (B) · P. Halder (B) Department of Computer Science & Engineering, University Institute of Technology, The University of Burdwan, Golapbag (North), Burdwan 713104, West Bengal, India e-mail: [email protected] P. Halder e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_6

67

68

S. Gupta and P. Halder

or behaviour prediction. Behaviour mining offers an umbrella under which several related research areas can converge. Human behaviour is not restricted to shopping trends or web usage alone, instead topics like fraud detection, traffic trends, moving patterns of stock prices, etc. can also be clubbed under the canopy of behaviour mining as each one of them can represent a unique behaviour of any user or a person. This paper deals with the domains of text mining and text analytics in order to build a system which uses a hybrid lexicon-based approach by utilising two different lexicons to consequently predict the sentiment and behaviour of an author. The two lexicons that we have used for this purpose are SenticNet 4.0 Lexicon (which comprises different sentiment words and whose each tuple indicates word, degree and rating of each word) and our own manually created Profile Lexicon. The rest of the paper is organised as follows: Sect. 2 discusses the previous related works of different researchers. Section 3 presents an overview of text mining. In Sect. 4, we have discussed the theories behind behaviour mining with an explanation on various models, classifications, characteristics and causes of human behaviour. Section 5 presents our proposed methodology in terms of an algorithm. Section 6 shows the datasets used by our system and the results obtained. Section 7 concludes the paper by putting across some future scope of improvements followed by references in the end.

2 Previous Related Work There are numerous research papers that have been published in the domain of sentiment analysis. We have selected a few popular ones with a special emphasis on those works where behaviour modelling is integrated. Paper [1] presents the idea of computer-assisted authorship attribution that is based on character-level n-gram language models. Here, the method is based on simple information theoretic principles, and achieves improved performance across a variety of languages like Greek, English and Chinese without using extensive pre-processing or feature selection. The proposed approach achieves 18% accuracy improvement over the best-published results for a Greek data set. In papers [2, 3], the authors have discussed multiple aspects of review text, such as grammatical, lexical, semantic and stylistic levels to identify important textual features. The features show a striking inclination and dependence on past reviews. The self-disclosed identity helps in identifying the reviewers that are displayed next to a review. Using random forest-based classifier, the authors have predicted the impact of reviews on sales. The three main feature categories discussed are ‘reviewerrelated’ features, ‘review subjectivity’ features and ‘review readability’ features. The papers show an integration of econometrics, text mining and predictive modelling techniques. Paper [4] discusses the customer’s behaviour using web mining techniques and its applications. The concept of web mining includes source data collection, data preprocessing, pattern discovery, pattern analysis and cluster analysis. As conventional

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System

69

methods are not appropriate for business situations and for finding out customers’ behaviours, customer segments are clustered by using k-means algorithm where input data is taken from web log of various e-commerce websites. Paper [5] analyses the log of online questions and chat messages which are recorded through a live video streaming system. The authors have used data mining and text mining techniques for analysing two different datasets and then they have performed an in-depth correlation analysis. The study revealed the discrepancies as well as similarities in the students’ patterns and themes of participation between online questions (student–instructor interaction) and online chat messages (student–students interaction or peer interaction). The relation between the number of online questions which were asked to the students and the students’ final grades so obtained highlighted the disciplinary differences in the online students’ participation. The work presented in [6] focuses on how to implement the event-based human behaviour recognition system. The proposed recognition procedure is divided into four subtasks, viz. video pre-processing, feature extraction, human behaviour recognition and resultant analysis. Artificial neural networks (ANNs) have been used for feature extraction. Matlab has been used for the purpose of implementation and various existing classifiers are implemented using the Weka suite.

3 Overview of Text Mining A. Definition Text mining is simply a method for drawing out content based on meaning and context from a large body of text. In other words, it is a method for gathering structured information from unstructured text. B. What is Text Classification? Text classification aims to assign predefined classes to text documents. It has been broadly studied in different communities such as data mining, database systems, machine learning and information retrieval. Text classification finds its applications in myriad domains, viz. medical diagnosis, image processing, document processing, etc.

4 Theories Related to Behaviour Mining Behaviour can be defined as the action or reaction of a person in response to an external or internal stimulus situation. We tend to understand about any person’s behaviour when we come to know about the real cause behind a person’s activity and/or the reason which had made the person behave that way. In case of behaviour mining, both understanding and evaluation are common reactions that any individual

70

S. Gupta and P. Halder

engages in on a daily basis. Psychology is a science of activity of people which leads to an understanding of the nature of behaviour. Behaviour is always the product of two things—the first one is the nature of an individual or an organism that behaves and the second one is the nature of a situation in which the individual finds himself/herself. Situation is a source of stimulus and is considered as the organism’s response to stimulus from the environment [7]. A. Models of Human Behaviour Several authors have understood about the notion of human behaviour from different mindsets and have proposed the human behaviour models differently. The two most prominent behaviour models [8] known in the literature are as follows: (a) Psychoanalytic Model: This model is based on the Freudian approach which focuses on the conflicting states of a human’s brain. As per the conflict model, Freud had used the clinical techniques of free association and psychotherapy to prove that it is not always possible to explain about behavioural changes with a conscious mind. He had further specified that unconsciousness is the major factor that governs the behavioural patterns of a person. Freud felt that a person’s behaviour generally depends on three factors, viz. Id, Ego and Superego. Id: In the context of human behaviour, the term ‘Id’ means pleasure. Although the possession of Id is considered to be a positive sign as it symbolises urge, desire and competitive nature in any human being, it also generates negative traits in an individual like aggression, prudence, dominance, ill-temperament, etc. It can make a person quarrelsome and incite in him/her the willingness to destroy. These traits are mostly found in young children. Ego: The term ‘ego’ denotes the ‘conscious’ stage in a person’s behaviour. The terms ‘Id’ and ‘ego’ conflict with each other, whereas the terms ‘ego’ and ‘superego’ go hand in hand. Superego: The word ‘superego’ denotes ‘conscience’, which is an inherent feature of any person and whose working is neither known to an individual nor controllable. It is dependent on two factors—the individual’s cultural values and the morals of the society. The development of conscience in an individual is greatly influenced by the activities and mannerisms of parents and family members. A child tends to unconsciously inculcate the ethics, values and morals from the people in his surroundings, especially from parents. There is always a tussle going on among Id, ego and superego and the degree of each of them varies from one person to another. Thus, the psychoanalytic model can prove to be useful in deciphering the variations in an individual’s personality traits like Id, ego and superego. Although this model is not supported by any empirical facts or figures and has always been criticised by modern authors, yet the way it deals with the theories of consciousness and unconsciousness can provide potential insight into a human being’s behavioural changes. (b) Existential Model: The existential model is based purely on the notion of literature and philosophy and has no scientific inclination. This model emphasises on the fact that because of the depersonalising effects of the surroundings, a

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System

71

person seeks to create his own mark by fighting all odds and by shaping his/her future. Due to the fast-paced lifestyles and the spread of urbanisation, people have become busy, self-centred and materialistic and have no time to adhere to traditional protocols. The primary focus of an individual is to make his/her life meaningful and worth living. Existential model is especially true for employees’ who feel the pressure to perform their best and sustain in the ongoing rat race going around in the market. B. Classification of Human Behaviour To analyse and measure behaviour, psychologists have divided behaviour into different classes. The following shows the various classifications of human behaviour [7]: (a) Overt and Covert Behaviour: Overt behaviour denotes those activities that occur externally and are observable, e.g. singing a song, dancing, playing an outdoor game, etc. denote overt behaviour. Covert behaviour, on the other hand, is not visible as it occurs inside a human body, e.g. thinking, understanding, analysing, etc. (b) Voluntary and Involuntary Behaviour: The behaviour that depends on the need and necessity of a human being is called voluntary behaviour. We can control and/or alter such a behaviour through learning and training, e.g. reading, writing, speaking, etc. Contrary to it, we have involuntary behaviour that occurs naturally and cannot be controlled by us, e.g. circulation of blood in the body, respiration, etc. (c) Molecular and Molar Behaviour: Molecular behaviour is driven by the sudden impulse or stimulus of a human being that does not incorporate any thought process, e.g. Mr. A happens to touch a hot iron and immediately moves his hands away on feeling the hot sensation. Molar behaviour, which is basically the opposite of molecular behaviour, involves a person to think and make a decision before acting on a situation, e.g. attacking an enemy after drafting a plan and/or strategy. C. Attributes of Human Behaviour We know that behaviour is an activity that can be observed, recorded and/or measured. Human beings respond to different situations in a variety of ways based upon their own personality traits and characteristic (attributes). Some of the popular characteristics of human behaviour [7] are as follows: (a) Social Rules and Regulations: As human beings live in a society, they are closely bound by social responsibilities, rules and regulations that all social beings have to follow for an amicable and healthy livelihood. (b) Language and Understanding: Language is a medium for sharing ideas and thoughts. It is through a language (in understandable form) that a person can express his/her emotions or feelings to another. This conversational interaction helps in the exchange of information, both at individual level and at group level.

72

S. Gupta and P. Halder

(c) Education and Knowledge: Education forms an integral part in one’s life as it prepares one to take correct decisions, i.e. empowers one to differentiate between what is correct and what is not. It is through educational learning that one tends to create his/her knowledge base. Proper education helps one in building an apt knowledge base and aids one in possessing a perfect skill set. Thus, education and knowledge affect human behaviour to a great extent. (d) Adaptability: Human beings, in general, are known to display a remarkable feature of adapting successfully to the changes in the environment. Thus, adaptability is surely an important aspect of human behaviour. (e) Aspiration: It is a general human tendency to behave in a manner so as to achieve certain goals and/or desires. The aim to accomplish goals, the willingness to fulfil desires and the aspiration to be successful drive a human to behave accordingly and showcase goal-directed behaviour. D. Causation of Behaviour Human beings generally display their behaviour while interacting with any individual or a group and/or while responding to a particular situation. We are known to respond to stimulus situations, and so there is always a cause sequence or cause–effect of human behaviour. This cause sequence or cause–effect relationship is commonly governed by the following parameters [7]: (a) Stimulus Situation: The situations where one person interacts with another person or the input a person receives in the form of sensations lead into the generation of impulses (or responsive pulses). Due to the environmental setup and/or activities of the surrounding entities, a human being reacts and showcases his/her behaviour. Behaviour may be any physical movement, a change in facial expression, pondering or introspection, etc. (b) Organism: It is said that whenever any stimulus situation occurs, then an organism gets started automatically. An organism is considered to be based on genetics, maturity, biological needs of a person, etc. It revolves around multiple learning scenarios involving personal demands, skills, attitude, ethics, etc. (c) Accomplishment: The feeling of accomplishing something causes a change in human behaviour. Events like accident, survival, winning, losing, etc. tend to change the behavioural outlook of an individual.

5 Our Proposed Methodology SenticNet 4.0 dataset (available both as a standalone XML repository and as an API2) is a well-known dictionary meant for the purpose of sentiment analysis. Let us understand about few notions of the SenticNet dataset and other such datasets.

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System

73

A. Performing Polarity Detection with SenticNet [9] Many researches use SenticNet as a sentiment lexicon because it offers the knowledge base that can be used to detect polarity from textual patterns in conjunction with Sentic patterns. Sentic patterns are sentiment-specific linguistic patterns that allow affective information to flow from one concept to another governed by the dependency relation between clauses and infer polarity of the text. This implements a rudimentary type of semantic processing, where the ‘meaning’ of a sentence is reduced to only one value, i.e. its polarity. B. Lexicons for Sentiment Analysis Sentiment analyser makes use of three sentiment lexicons, viz. SentiWordNet, SenticNet and SentiSlangNet to find the polarity of each text and use this information to generate cliques based on the sentiments on each issue. The folloswing presents a detailed discussion on the various lexicons [10, 11] used for performing sentiment analysis: (a) SentiWordNet: SentiWordNet is a domain-independent lexical resource used for performing sentiment analysis. It is distributed freely for the purpose of research. SentiWordNet is derived from WordNet by assigning positive, negative and objective scores to all WordNet synsets. Each sentiment score is a real value that lies in the interval [0, 1] and signifies the positive, negative and neutral values of each term contained in the synset. The number of terms in the SentiWordNet is very high compared to other lexicons. There are around 1,17,659 senti-synsets available in SentiWordNet that is equivalent to the number of synsets available in WordNet. But as SentiWordNet contains only uni-grams, it can provide sentiment scores only at the syntactic level. (b) SenticNet: SenticNet is a public lexical resource created using the notion of Sentic computing and can be viewed as a knowledge base to analyse social data. Sentic is a term derived from Latin words ‘sentire’ and ‘sensuous’. The domain of Sentic computing deals with common sense and effective computing for analysing documents with varying granularity from paragraph to clause level apart from analysing opinions over the web. Further speaking, SenticNet 2 provides polarity of each concept and also its Sentic values along with its top 10 affectively related concepts. (c) SentiSlangNet: Social media texts such as tweets, posts, comments, etc. are rich in emoticons, slangs and abbreviations. Further, the slangs used in social media texts are evolving rapidly. Most of the current approaches have not considered the sentiment score of slangs while analysing the sentiment of the user. SentiSlangNet is a sentiment lexicon for slangs that is created by web scraping slangs from the web. C. SenticNet 4.0 Dictionary The SenticNet 4.0 dictionary comprises a score for each concept linked with each synset based on the emotional label. In the same for-loop, each word is passed to SenticNet for calculating a polarity score which is in the range of

74

S. Gupta and P. Halder

[−1, 1], where −1 denotes a very negative emotion and 1 denotes a very positive emotion. In case the word is not found in the SenticNet 4.0 dictionary, then that iteration of the loop gets skipped. For every iteration that is successful and gets completely executed, a countervariable is increased by one and the polarity score of the word is added to a running total polarity score sum. The total polarity score sum is then divided by the value of the counter for generating the sentiment score for the article. This process gets repeated for each of the texts available in a single day. The average sentiment score of the entire text is accepted as the daily average sentiment score for an entity [12]. D. Profile Lexicon The Profile Lexicon is a self-made dictionary that consists of 20 different behaviour and/or characteristics classes. Each of these classes comprises sentiment words denoting the emotion as described by the corresponding class. For instance, the class ‘Sadness’ contains sentiment words like sadness, unhappy, sorrowful, dejected, regretful, depressed, downcast, miserable, downhearted, desolate, etc. We have inserted all the nouns, adjectives, verbs and adverbs with the matching sentiment value of words under each behaviour class. Table 1 shows the different behaviour classes that we have manually created to form our Profile Lexicon. E. Our Proposed Algorithm We have proposed an algorithm that takes in unstructured text and displays the sentiment score, sentiment and profile weight matrix using two lexicons, viz. SenticNet 4.0 Dictionary and our own Profile Lexicon. The algorithm is as follows: Step 1: Step 2:

Take some raw data and store this data into a database. Preprocess the data. In this step, tokenisation takes place. The tokens are then stored in a database (Structured Database). Step 3: The tokens are fetched and checked with the contents of the SenticNet Database one by one until all the tokens are analysed. Step 4(a): If the word is found in SenticNet Database then the degree and rating of the matched token are retrieved and/or returned. Step 4(b): If the word is not found, then the next token is fetched. Then go to Step 3. Step 5: If the word is present, then calculate the SentiScore using the formula: SentiScore = SentiScore + Rating. Table 1 Different behaviour classes of our Profile Lexicon Sadness

Disappointment

Joy

Anger

Romance

Drama

Inspiration

Crime

Fear

Disgust

Dislike

Like

Surprise

Anticipation

Frustration

Confusion

Sympathy

Peace

Excitement

Nature lover

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System

75

Step 6(a): Next, check whether the token is present in the Profile Lexicon or not. If the token is found and matched, then retrieve the matching profile. Then calculate the profile weight and check with the next token. Step 6(b): If the word is not found to the Profile Lexicon then update the Profile Lexicon manually. Step 7: Display SentiScore, Sentiment and Profile Weight matrix.

6 Implementation and Results Datasets: Here we have used SenticNet 4.0 as a training dataset along with our own Profile Lexicon. Using these datasets, we can easily predict whether any entered word in a sentence is positive or negative, calculate the sentiment score of each positive and negative word, calculate the overall sentiment score of the whole test data and predict the behaviour of the author. Our Profile Lexicon contains about 3500 sentiment words clubbed under 20 different behaviour classes. We have considered several online sources like Twitter, review forums, blogs, etc. for building our Profile Lexicon. Then we have run 300 sentences on our system for the purpose of testing. Result and Analysis: The following tables—Table 2 shows the confusion matrix for the classes Sentiment = Positive and Sentiment = Negative, Table 3 shows the performance measures of Sentiment Prediction and Table 4 shows the performance measures of Behaviour Prediction. Table 2 Confusion matrix for the classes Sentiment = Positive and Sentiment = Negative Predicted class

Actual class

Sentiment = Positive

Sentiment = Negative

Total

Sentiment = Positive

137 (Tp)

13 (Fn)

150 (P)

Sentiment = Negative

18 (Fp)

132 (Tn)

150 (N) 300 (P + N)

Table 3 Performance measures of sentiment prediction

Measure

Formula

Result (%)

Accuracy

(Tp + Tn)/(P + N)

89.67

Error rate

(Fp + Fn)/(P + N)

10.33

Recall

Tp/P

91.33

Precision

Tp/(Tp + Fp)

88.39

F-score

(2*Precision*Recall)/(Precision + Recall)

89.84

76 Table 4 Performance measures of behaviour prediction

S. Gupta and P. Halder Behaviour prediction

Accuracy (%)

Error rate (%)

Correct

271

90.33

9.67

Incorrect

29 300

Hence, we can conclude that our proposed system tends to offer a fair and acceptable accuracy of 89.67% while detecting and predicting sentiments. The accuracy obtained for predicting behaviour is 90.33% which is better than that of sentiment prediction, thereby showcasing the efficiency and robustness of our Profile Lexicon.

7 Future Scope and Conclusion In this paper, we have presented an approach to perform sentiment analysis on texts so as to evaluate the sentiment score and sentiment as conveyed in the text. Along with it, we have built our own lexicon named Profile Lexicon for predicting the behaviour and/or characteristic of the author from his/her inputted texts. As for future course of action and endeavour, we aim to enhance the accuracy of our system by incorporating benchmark datasets and by using an exhaustive list of words and dictionary entries. We will also try to improve our algorithm by modelling more complex syntactic and semantic structures.

References 1. Peng F, Schuurmans D, Wang S, Keselj V (2003) Language independent authorship attribution using character level language model. In: Proceedings of the tenth conference on European chapter of the association for computational linguistics, vol 1, Budapest, Hungary. ACM, pp 267–274 2. Ghose A, Ipeirotis PG (2008) Estimating the socio-economic impact of product reviews: mining text and reviewer characteristics. NYU Stern research working paper 3. Ghose A, Ipeirotis PG (2010) Estimating the helpfulness and economic impact of product reviews: mining text and reviewer characteristics. IEEE Trans Knowl Data Eng 23(10):1498–1512. IEEE 4. Yadav MP, Feeroz M, Yadav VK (2012) Mining the customer behaviour using web usage mining in e-commerce. In: Third international conference on computing communication and networking technologies (ICCCNT), pp 1–5, Coimbatore, India 5. He W (2013) Examining students online interaction in a live video streaming environment using data mining and text mining. J Comput Hum Behav 29(1):90–102. Elsevier 6. Lu J (2016) Empirical approaches for human behaviour analytics. School of Engineering, Computing and Mathematical Sciences 7. Textile Apex webpage https://textileapex.blogspot.com/2014/08/human-Behaviour.html. Accessed 28 July 2018 8. Tiwary V (2012) Social processes and behavioural issues. Gullybaba Publishing House (P) Ltd

A Hybrid Lexicon-Based Sentiment and Behaviour Prediction System

77

9. Cambria E, Poria S, Bajpai R, Schuller B (2016) A semantic resource for sentiment analysis based on conceptual primitives. In: Proceedings of the 26th international conference on computational linguistics: technical papers (COLING 2016), pp 2666–2677, Osaka, Japan 10. Rafeeque PC (2014) Large scale short text analysis in Twitter to identify same wavelength communities. Faculty of Information and Communication Engineering, Anna University 11. Rafeeque PC, Sendhilkumar S, Mahalaxmi GS (2014) Twitter sentiment analysis for large-scale data: an unsupervised approach. Cogn Comput 7(2):254–262. Springer 12. Sorto M, Aasheim C, Wimmer H (2017) Feeling the stock market: a study in the prediction of financial markets based on news sentiment. In: Proceedings of the southern association for information systems conference (SAIS 2017), St. Simons Island, GA, USA, AISeL

Object Detection in Clustered Scene Using Point Feature Matching for Non-repeating Texture Pattern Soumen Santra, Partha Mukherjee, Prosenjit Sardar, Surajit Mandal and Arpan Deyasi

Abstract Effective object detection must be able to handle cluttered visions which convert into the object size, location, orientation, and other movements. We presumed that Computer Vision System Toolbox™ MathWorks offers a variety of techniques for handling challenges in object detection. In this paper, we elaborate on how to detect an object in a cluttered scene, given a reference image of the object. The output of this paper explains an algorithm for detecting a recognized object depending on finding the vision points correspondences between reference and target images. It can detect each and every object in spite of a scale change or in-plane rotation and quite extend to robust with small amounts of out-of-plane rotation. This method of object detection through recognized feature points works best for objects that exhibit nonrepeating texture patterns, which give rise to unique feature matches. In connection with this, present algorithm is designed for detecting a specific static object only. Keywords Object detection · Shape invariant · Reference image · Vison points · In-plane rotation

S. Santra (B) · P. Mukherjee · P. Sardar Department of Computer Application, Techno International New Town, Kolkata, India e-mail: [email protected] P. Mukherjee e-mail: [email protected] P. Sardar e-mail: [email protected] S. Mandal Department of Electronics and Communication Engineering, B P Poddar Institute of Management and Technology, Kolkata, India e-mail: [email protected] A. Deyasi Department of Electronics and Communication Engineering, RCC Institute of Information Technology, Kolkata, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_7

79

80

S. Santra et al.

1 Introduction For two overlapped images, image splice technology is nowadays commercially used through matching of the coordinate points. This technique works as spatial transformation, which is broadly classified in feature-based matching [1] and area-based matching [2]. Contrary to statistical learning techniques, the former one is attractive for researchers, which is primarily analyzed on non-repeating texture patterns. Due to high robustness, smaller fluctuation in the end result due to brightness variation in the image and faster real-time processing, this method is more significant, but computation time is delayed when normalized cross-correlation technique [3] is invoked. Ommer proposed a hierarchical technique [4] based on information collections from the edge of the boundaries, and candidates are further refined at the verification stage. But this technique is mostly applicable for static object detection. Nagase and coworkers developed a novel object matching technique using feature point detection where attribute values are considered to minimize statistical error [5]. Novel stereo matching technique [6] is proposed by Yan using SIFT feature extracting algorithm though there was a lack of detailed texture. Harris et al. explained Corner Detector and Fast Retina Keypoint detector which are used as object detection and identification [7] with very good robustness. In presence of in-plane rotation, detection technique is established by Reddy et al. [8]. The same is also calculated by Sravani [9], which is considering the out-of-plane rotation. Bodke applied this method for uniformly colored objects [10]. Method is also used for deformable shape [11] when rigid boundary points are not available. For symmetric part detection, compact superpixel image is generated [12] using medial axis. Quality and reliability for feature point detection [13] are introduced by Ratanasanya et al. and claimed as a better algorithm than SIFT. Novel foreground detection associated with background modeling algorithm is proposed for low-memory requirement [14]. In this present paper, object detection is made in cluttered environment from reference image even in presence of in-plane rotation, and robustness is verified for small amount of out-of-plane rotation. Work is carried out for non-repeating texture patterns using computer vision system toolbox [15]. There is no such work where in-plane and out-plane rotations are both applied together in a method. But here we can show our method both matched in-plane and out-plane region of the object with the original image.

2 Algorithm A brief description of the algorithm on which the work is carried out is provided in this section upon which results are obtained and represented in Sect. 3. At first, both source and object images are converted into grayscale image from RGB scale, and their surface features are individually detected. Feature descriptors at different insertion points are identified and corresponding transformations are calculated. Once

Object Detection in Clustered Scene Using Point Feature …

81

the matching point pairs are identified, they are located with the outliers of the object. Data polygon is created, and it is transformed into coordinate system of the source image, thereafter located as the matched image of object. Step 1: Step 2: Step 3: Step 4: Step 5: Step 6: Step 7: Step 8: Step 9: Step 10: Step 11: Step 12: Step 13: Step 14:

Read Source Image. Convert it into GrayScale Image from RGBScale. Read Object Image to be detected from Source Image. Convert it into GrayScale Image from RGBScale. Detect the surface feature of the GrayScale Source Image. Detect the surface feature of the GrayScale Object Image. Plot the surface feature Points for both of the Images. Extract feature descriptors at the interest points in Source Image and Object Image. Match the feature descriptor using their descriptors. Locate putatively matched feature descriptor. Calculates the transformation relating the matched points, while eliminating outliers to localize the object. Locate the matching point pairs with the outliers removed figure. Create a data-type Polygon and transform the Polygon into Coordinate System of the Source Image. Locate the Polygon as Matched Object Image from the Source Image.

3 Methodology of Object Detection A. Step 1: Reading the Images: We have an image of a cluttered scene, i.e., the object that to be detected in *.jpg format, and we want to detect that particular object for which we have a separate image, the original image file in .jpg format. We can start by reading the original image containing the object of interest. Here we take the original image (Figs. 1 and 2) and convert it into a grayscale image (Fig. 3) by browsing it from our computer. Then we take the object of interest, i.e., the image which we want to detect from that original image and convert that into gray scale. After converting into gray scale, we saved it into directory and crop section is identified, as shown in Figs. 4 and 5, respectively. B. Step 2: Detecting Feature Points: We can detect feature points in both images. Then, we can visualize the strongest feature points found in the reference image, depicted in Fig. 6. C. Step 3: Extracting Feature Descriptors: We can extract feature descriptors at the interest points in both images in Figs. 7 and 8.

82

Fig. 1 Select images from photo diary [16]

Fig. 2 Select the image into panel [16]

S. Santra et al.

Object Detection in Clustered Scene Using Point Feature …

Fig. 3 Convert the original image into grayscale image

Fig. 4 Save into directory of device

83

84

Fig. 5 Finding crop section as object from the original image

Fig. 6 Convert into gray image of detecting object

S. Santra et al.

Object Detection in Clustered Scene Using Point Feature …

Fig. 7 Extracting feature descriptor for crop section object

Fig. 8 Locating feature descriptor for whole image and finding putative point

85

86

S. Santra et al.

Step 4: Finding Putative Point If these point matches then match the features using their descriptors in Fig. 8. D. Step 5: Locating the Object in the Scene Using Putative Matches: The estimate geometric transform function calculates the transformation relating the matched points, while eliminating outliers. This transformation allows us to localize the object in the scene as in Fig. 9. Next, get the bounding polygon of the reference image. To indicate the location of the object in the scene, we can transform the polygon into the coordinate system of the target image (Fig. 10). Then, we can display the detected object shown in Fig. 11. Here we first explained the in-plane and out-plane regions of image with features and then show the robustness of approached method with features of image rotations.

Fig. 9 Locating the object in the scene using putative matches for outline points

Fig. 10 Locating the object in the scene using putative matches for inline points

Object Detection in Clustered Scene Using Point Feature …

87

Fig. 11 Boundary polygon for matched detecting object in the scene

All the figures related to output show those parameters. There are no such algorithms which worked on both the plane rotations but our approach works on both in-plane and out-planes regions because putative matches is such a feature point which works throughout the object along all vector planes, which proved through the output images shown in (Tables 1 and 2).

88

S. Santra et al.

Table 1 Robustness about inlier plane and outlier plane In-plane region

Out-plane region

[17]

In-plane region—the image rotates in the image plane

Out-plane region—the image

Spin the axis along the vector through the plane

Spin the axis out of the plane

Any other rotation where it only shows that object has rotated through axis and nothing else changes

Any other rotation where the new position of the vector is not in its last plane

In-plane translation truly performed

Superior–inferior truly performed

rotates out of the image plane

Left–right truly performed Anterior–posterior truly performed Entropy–superpixel ratio truly performed

Flipping on image truly performed

Angle compensation truly performed

Blending on image truly performed

Non-maximum suppression and validation of post-processing truly performed for both Shape features and appearance features both truly performed same for both [12] shown in Figs. 7 and 8 Agglomerative clustering for detect object truly performed with source image for both [12] shown in Figs. 9 and 10 Table 2 All putative points’ pictures of Object Image on rotations

Object Detection in Clustered Scene Using Point Feature …

89

In the next two sections (Sects. 4 and 5), we have made a comparative study on robustness of in-plane and out-plane, and corresponding putative point pictures of rotated object images are provided. Details of object image with matching property are provided in Table 3, whereas translation and shearing are represented in Tables 4 and 5, respectively.

4 Robustness of Method for In-Plane and Out-Plane Region of Image Here we first explained the in-plane and out-plane regions of image with features and then show the robustness of approached method with features of image rotations. All the figures related to output show those parameters. There are no such algorithms which worked on both the plane rotations but our approach works on both in-plane and out-plane regions because putative matches is such a feature point which works throughout the object along all vector planes, which is proved through the output images Figs. 9 and 10.

5 Analysis of Results See Table 2.

6 Conclusion Using pattern matching and feature point mapping classification, we can recognize image or detect the object from the target image reference. In most of the cases, from the existing work on object recognition or object detection, all feature point matching could not be possible. But here we can show mapping between all feature points of source and target image sets. We also introduced the methodology for selecting small image set from the target one. Here all the matched point of image sets are determined through in-plane as well as might be out-plane rotations. This technique also represents a shape descriptor or classifier with the determined feature points. Here, through the output images, we can also detect the object including both the in-plane and out-plane regions of the object with the main source object.

Image

5

500

500

500

5

10

15

Putative points Image

0

Rotation degree

Table 3 Details of Object Image on rotations Inlier image

3

5

6

6

No. of points

Outlier image

3

6

10

12

No. of points

One object from different plane including 2 objects

2 objects properly

2 objects properly

2 objects properly

No. of objects matched

(continued)

Crop but Not proper. Matched 2 different objects from different planes

Crop properly

Crop properly

Crop properly

Remarks

90 S. Santra et al.

Image

Table 3 (continued)

Putative points Image

500

500

500

Rotation degree

20

25

30

No inlier form

Inlier image

3

3

0

No. of points

Outlier image

4

5

2

No. of points

Inlier proper but in outlier matched 2 different objects from 2 different planes

Inlier proper but in outlier matched 2 different objects from 2 different planes

No objects

No. of objects matched

(continued)

Crop proper where inlier right but outlier 2 objects found from 2 different planes

Crop proper where inlier right but outlier 2 objects found from 2 different planes

Not crop

Remarks

Object Detection in Clustered Scene Using Point Feature … 91

Image

Table 3 (continued)

Putative points Image

500

500

500

Rotation degree

35

45

55

Inlier image

3

4

3

No. of points

Outlier image

4

5

5

No. of points

2 objects properly

2 objects properly

2 objects properly

No. of objects matched

(continued)

Crop properly

Crop properly

Crop properly

Remarks

92 S. Santra et al.

Image

Table 3 (continued)

Putative points Image

500

500

500

Rotation degree

65

75

90

Inlier image

7

5

3

No. of points

Outlier image

12

8

5

No. of points

2 objects properly

2 objects properly

1 objects properly only

No. of objects matched

Crop properly

Crop properly

Crop only one object

Remarks

Object Detection in Clustered Scene Using Point Feature … 93

Image

500

Putative points image

Inlier image

Table 4 Details of Object Image on (15, 25) pixel-based translation

3 But Matched 2 different planes including match object

No. of points

Outlier image 4 But Matched 3 different planes including match object

No. of points

Two objects not properly matched. Middle of two objects matched in the crop section

No. of objects matched

Crop middle from the two objects

Remarks

94 S. Santra et al.

Image

Putative points image

500

500

500

500

Shear details

X half

X twice

Y half

Y twice

Table 5 Details of object image on shearing along X- and Y-axis

No

No

No

No

Inlier image

0

0

0

0

No. of points

No

No

Outlier image

1

1

0

0

No. of points

No match

No match

No match

No match

No. of objects matched

No crop

No crop

No crop

No crop

Remarks

Object Detection in Clustered Scene Using Point Feature … 95

96

S. Santra et al.

References 1. Lowe DG (2004) Distinctive image features from scale-invariant key points. Int J Comput Vision 60(2):91–110 2. Reddy BS, Chatterji BN (1996) An FFT-based technique for translation rotation, and scaleinvariant image registration. IEEE Trans Image Process 3(8):1266–1270 3. Bojiao D, Donghua Z (2007) Fast matching method based on NCC. Transducer Microsyst Technol 26(9):104–106 4. Ommer B, Malik J (2009) Multi-scale object detection by clustering lines. In: 12th International conference on computer vision 5. Nagase M, Akizuki S, Hashimoto M (2013) 3-D feature point matching for object recognition based on estimation of local shape distinctiveness. In: International conference on computer analysis of images and patterns, pp. 473–481 6. Yan Y, Xia H, Huang S, Xiao W (2014) An improved matching algorithm for feature points matching. In: International conference on signal processing, communications and computing 7. Ben-Musa AS, Singh SK, Agrawal P (2014) Object detection and recognition in cluttered scene using Harris Corner Detection. In: International conference on control, instrumentation, communication and computational technologies 8. Reddy KR, Krishna KVS, Kumar VR (2014) Detection of objects in cluttered scenes using matching technique. Int J Electron CommunTechnol 5(3):42–44 9. Sravani C, Harikrishna B, Gayatri K, Anusha K, Pydiraju K (2015) Object capturing in a cluttered scene by using point feature matching. Int J Eng Res Appl 5(3):49–52 10. Bodke VS, Vaidya OS (2017) Object recognition in a cluttered scene using point feature matching. Int J Res Appl Sci Eng Technol 5(IX):286–290 11. Patil S, Patil NC (2015) Object localization using putative point matching in cluttered scene. J Emerg Technol Innov Res 2(6):3088–3092 12. Lee T, Fidler S, Levinshtein A, Sminchisescu C, Dickinson S (2015) A framework for symmetric part detection in cluttered scenes. Symmetry 7:1333–1351 13. Ratanasanya S, Polvichai J, Sirinaovakul B (2015) Feature point matching with matching distribution. Recent Adv Inf Commun Technol 9–18 14. Tsai WK, Sheu MH (2016) An efficient foreground object detection method using a color cluster-based background modeling algorithm. In: International symposium on computer, consumer and control 15. https://www.mathworks.com/products/computer-vision.html 16. Random photography (2016) Place: Ultadanga, Golaghata, camera model: Nikon D90, resolution: 4288 × 2848 (12.3 effective megapixels) edited with snapseed courtesy by: Partha Mukherjee 17. Santra S, Mandal S (2018) A new approach towards invariant shape descriptor tools for shape classification through morphological analysis of image. In: 2nd International conference on computational advancement in communication circuit and system

Human Behavior Recognition: An l 1 – l s KSVD-Based Dictionary Learning and Collaborative Representation-Based Classification Pubali De, Amitava Chatterjee and Anjan Rakshit

Abstract This work presents a new idea for human behavior recognition based on dictionary learning algorithm and collaborative representation-based classification approach. In this paper, we have proposed an l1 – l s -based KSVD algorithm for learning a dictionary and collaborative representation is used in the classification phase for this problem. The performance of our proposed idea for human behavior recognition problem establishes the superiority of our new idea. Keywords Dictionary learning · Human behavior recognition · CRC · l1 – l s

1 Introduction The human action recognition is the most exciting research problem [1, 2] in the modern world as well as the successful application of sparse representation and machine learning theory in the field of face recognition [3], image classification, and [4] image denoising [5] which invites the attention of researcher to implement the sparse representation in the field of action recognition problem. The dictionary learning technique based on sparse representation already established its effectiveness over a fixed dictionary-based representation to find out the desired representation of a signal [6]. In the case of dictionary learning-based sparse representation implies that better representation of a signal is trusting on the feature of learned dictionary. Original KSVD dictionary learning-[7, 8] based human behavior recognition had already P. De (B) Electrical Engineering Department, Techno International Batanagar, Kolkata, India e-mail: [email protected] A. Chatterjee · A. Rakshit Electrical Engineering Department, Jadavpur University, Kolkata, India e-mail: [email protected] A. Rakshit e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_8

97

98

P. De et al.

implemented in our earlier paper [6] and the executed results prove its utility for action recognition problem. This success of dictionary learning algorithm influences us to present a new idea based on KSVD dictionary learning [7] to get better recognition accuracy. As we all know, the learning of a dictionary using basic KSVD [7, 8], is commonly executed by following two stages which are sparse coding stage and dictionary updating stage. Usually, basic KSVD-based classification procedure [6] solves the sparse coding stage as a l 0 minimization problem and sparse representation-based classifier (SRC) is used for the classification stage [3]. We have used this basic KSVD- and SRC-based classification for human behavior recognition to solve l0 minimization problem and orthogonal matching pursuit algorithm [8] is used from greedy family and it provides the satisfactory result [6]. Nowadays, due to the major application of l1 minimization [9, 10] for an underdetermined system to find out the sparsest solution draws the researcher attention to apply this in the field of face recognition [11], compressive sensing [12],etc. Here, keeping this feature of l1 minimization [9, 10] in mind, we have implemented this in a sparse coding stage for finding the sparsest solution in dictionary learning algorithm by solving some underdetermined system considering some measurement error in linear equation, i.e., Y = Dx + e and minimization of problem becomes min x x1 subject to y − Dx ≤ ε There are lots of methods to solve the above minimization problem [13–16]. l1 minimization method with including regularization parameter (l1 –l s )becomes a most interesting idea for signal representation [10], we have solved this above problem as l1 regularized least square problem by truncated Newton interior-point method [13]. After learning the dictionary D, we have implemented collaborative representation (CR)-based SRC [17] for classification purpose to implement the advantage of CR as it can identify an unknown class from a larger set of training dataset including all the event classes and the resulting analysis has produced better classification accuracy and established the effectiveness of our proposed l 1 –l s KSVD-based CRC [17] classification approach and produces better result over SRC [5, 10]. The presentation of this paper is embodied as follows. Section 2 presents l 1 –l s based KSVD dictionary learning algorithm. Section 3 provides CRC-based classification approach. Section 4 presents the resulting analysis and produce the effectiveness of our approach over the conventional methods. Section 5 provides the conclusion of this work.

2 l 1 – l s KSVD-Based Dictionary Learning The dictionary learning [7, 8] is one of the most encouraging ideas in recent days which is used numerously for image classification purposes and this dictionary learning-based sparse representation approach has outperformed in our previous paper [6] over other fixed dictionary-based approaches.

Human Behavior Recognition: An l 1 – l s KSVD-Based Dictionary …

99

Inspiring by the performance analysis [6], in this paper we have proposed l1 – l s -based KSVD algorithm to find the sparsest solution for an underdetermined system in place of commonly used greedy algorithm [8] and this l 1 – l s [10, 13] includes the regularization parameter with normal l1 minimization problem using truncated Newton interior-point method [13] which gives the essence of fast l 1 minimization [10]. The dictionary is learned in original KSVD dictionary [7, 8] by computing the two stages, i.e., 1) sparse coding stage and 2) dictionary update stage, simultaneously within an iterative loop. In our proposed approach, the sparse coding stage is performed as l1 – l s minimization[10, 13], the problem which is followed by conventional dictionary update stage. The following section describes the l1 – ls -based complete KSVD algorithm. In the case of KSVD dictionary learning algorithm [7, 8], usually we have used algorithm from greedy and relaxation family to solve the l0 minimization problem in sparse coding stage. Here, we have cast this l1 regularized least square problem for finding sparse solution and it converts the objective function as a convex problem and there are many methods [13–16] to solve the above problem. Here we have used interior point-based method [13] to solve this problem. This l1 regularized least square method combines the l1 minimization problem with regularization parameter [9, 10] as 1 min y − Dx22 + λx1 x 2

(1)

Here λ presents the regularization parameter and its value is λ > 0. Here, we have used this above least square problem combining it with l1 minimization to reduce computational burden by solving it using truncated Newton interior-point method (TNIPM) to find the best sparse solution algorithm and it is given in [10, 13]. The next phase of dictionary learning algorithm [7, 8] is to find out a small learned dictionary from a larger training dataset in such a way that a small block can well adapt the larger training dataset using the sparse vector obtained from l1 – l s minimization. Here, our work has shown that l1 – l s -based approach [10] can enhance the performance of small dictionary adaptation from larger training database. l1 – l s KSVD dictionary learning algorithm is given below in algorithm I.

100

P. De et al.

3 CRC-Based Classification The basic idea behind SRC [3]-based classification is to code the unknown test signal y based on the fixed matrix data matrix D by following the equation y = Dx where the sparse vector is x. The sparse vector can be found out by using l1 or l 0 minimization [3]. l1 minimization problem is more advantageous as it can handle over-complete dictionary easily and can provide the sparsest solution. But it takes more time to find sparse vector that can be solved by many fast l1 minimization problems [13–16] Here we have proposed l1 regularized least square problem [10, 13] to solve sparse solution.

Human Behavior Recognition: An l 1 – l s KSVD-Based Dictionary …

101

In the case of CRC [17], it is a special case of SRC where unknown test signal y is represented by using complete training database including all the classes. Our work presents that CR based SRC also give effective performance over l1 based SRC. It implies that not only sparsity but also CR-based signal representation [17] is more effective for human behavior recognition. In CRC [17], the test signal y is represented based on the learned dictionary D from KSVD algorithm by using l1 regularized least square method. The CR-based classification is obtained by using Eq. (1) and derived as −1  ρ = DT D + λ.I DT y

(2)

 −1 Assume Q = DT D + λ.I DT , this Q is calculated as the projection matrix and y is coded based on this projection matrix. After computing the value of Q, the operation of CRC [17] becomes equal to SRC classifier for testing the unknown sample class by class. The complete CR-based least square algorithm [17] is given in this work as a new idea for classification of human behavior over the conventional SRC method which has been successfully utilized in our previous paper [6] for action recognition problem. The CRC algorithm [17] based on l 1 – l s KSVD is given below.

4 Evaluation of Performance In this research work, we have presented our new idea of KSVD dictionary learning algorithm based on l 1 regularized least square [10, 13] as sparse coding stage and CRC-based classification process for action recognition problem. Here we have chosen the benchmark action recognition dataset [18] as our problem statement. This

102

P. De et al.

dataset is [18] composed of 700 trials of human behaviors taken by different persons. This dataset contains different kinds of human behavior like sitting down on a chair, standing up from a chair, getting up from the bed, pouring water to the glass, walking, and drinking a glass of water. These typical behaviors of human are identified based upon our new approach. Here, a small block from a larger training database as a form of a dictionary D is learned by using KSVD dictionary learning algorithm [7, 8]. The modification of original KSVD [7, 8] is done by using l 1 – l s method [10, 13] as it is more applicable for over-complete dictionary and produces more sparse solution. After learning the dictionary as D to adapt all the training signals, this dictionary D is used as input to CRC classifier [17] as a training dataset. Here we have replaced the CRC classification [17] in place of SRC to modify our original KSVD-based classification for human recognition problem [6] and this modified approach is tested over other approaches to justify the two reasons which are given below. (1) Application of l 1 – l s algorithm [10] in place of l 0 minimization in sparse coding stage to solve the sparse solution for dictionary learning algorithm and to produce the sparsest solution in place of a conventional greedy algorithm [9] like OMP which is normally used in sparse coding stage of KSVD [7, 8]. (2) Application of CR-based SRC [17] in place of conventional SRC-[3] based conventional classification method which implies that unknown test sample y can be obtained over a complete training database which includes all the classes and CRC algorithm [17] based on l1 regularized least square method reduces the computational burden and also reduces the complexity. Our proposed idea based on human behavior classification is compared with our previous approach [6] and with the l 1 – l s KSVD-based classification where dictionary learning and classification stage, in both the cases, l1 – l s method [10, 13] is used for finding out the sparsest solution. The resulting analysis produced a satisfactory result and provides the usefulness of our proposed idea. Table 1 shows the performance analysis of our new approach compared to other possible approaches; as a class classification problem, where we aim to find out the presence of unknown class is present to an exact class or not. The obtained result establishes that our proposed idea for human recognition problem produces a satisfactory result. This resulting analysis implies that CR-based SRC performs superior to the other conventional KSVD and l1 – l s -based approach for action recognition problem. This analysis also presents that solving the sparse coding stage as l1 regularized least square problem produces the best dictionary for classification stage which enhances the recognition accuracy for that same activity recognition problem set. Figure 1 presents the graphical representation of recognition accuracy for each behavior class and compared it with the conventional KSVD-based classification approach and l1 – l s -based classification approach. Here , method 1 presents original

Human Behavior Recognition: An l 1 – l s KSVD-Based Dictionary …

103

Table 1 Comparative analysis of recognition accuracy for human behavior recognition Dictionary learning stage

Classification stage

Behavior classes of human

Recognition accuracy (%)

OMP-based sparse coding

OMP-based SRC

Standup chair

69

Sit down chair

75.67

Pour water

94

Get up bed

69.99

l 1 - l s -based sparse coding

l 1 - l s -based Sparse coding

l 1 - l s -based SRC

CR-based SRC using l1 - ls

Drink glass

100

Climbing stairs

85

Walk

85

Standup chair

30

Sit down chair

45

Pour water

95

Get up bed

100

Drink glass

85

Climbing stairs

100

Walk

80

Standup chair

95

Sit down chair

100

Pour water

92

Get up bed

90

Drink glass

100

Climbing stairs

95

Walk

85

KSVD dictionary learning-based SRC approach. Method 2 denotes l1 – l s KSVDbased SRC classification approach, and method 3 presents l1 – l s KSVD-based CRC classification for human behavior recognition.

5 Conclusion This research paper introduces a new KSVD dictionary learning approach-based CRC classification for human behavior recognition problem. Here, sparse coding stage in dictionary learning algorithm is presented as l1 – l s problem for presenting best dictionary to adapt the total training dataset for classification of human behavior. The recognition accuracy acquired from our approach presents a satisfactory result compared to other approaches.

104

P. De et al.

Fig. 1 Graphical representation of recognition accuracy for each behavior class

References 1. Debes C, Merentitis A, Sukhanov S (2016) Monitoring activities of daily living in smart homes: understanding human behaviour. IEEE Signal Process Mag 33(2):81–94 2. Erden F, Velipasalar S, Alkar AZ (2016) Sensors in assisted living: a survey of signal and image processing methods. IEEE Signal Process Mag 33(2):36–44 3. Wright J, Yang A, Ganesh A, Sastry S, Ma Y (2009) Robust face recognition via sparse representation. IEEE Trans Pattern Anal Mach Intell 31(2):210–227 4. Yang J, Wang J, Huang T (2011) Learning the sparse representation for classification. In: Proceedings of IEEE International Conference on Multimedia Expo (ICME), July 2011, pp. 1–6 5. Elad M, Aharon M (2006) Image denoising via sparse and redundant representations over learned dictionaries. IEEE Trans Image Process 15(12):3736–3745 6. De P, Chatterjee A, Rakshit A (2018) Recognition of human behavior for assisted living using dictionary learning approach. IEEE Sens J 18(6):2434–2441 7. Aharon M, Elad M, Bruckstein A (2006) K-SVD: an algorithm for designing overcomplete dictionaries for sparse representation. IEEE Trans Signal Process 54(11):4311–4322 8. Elad M (2010) Sparse and redundant representations: from theory to applications in signal and image processing. Springer, Berlin 9. Donoho D, Tsaig Y (2006) Fast solution of l 1 norm minimization problems when the solution may be sparse (preprint) 10. Yang A, Ganesh A, Zhou Z, Sastry S, Ma Y (2010) Fast l 1 minimization algorithms and an application in robust face recognition: a review. Technical report UCB/EECS-2010-13, UC Berkeley 11. Bruckstein A, Donoho D, Elad M (2007) From sparse solutions of systems of equations to sparse modeling of signals and images. SIAM review (in press) 12. Hale E, Yin W, Zhang Y (2007) A fixed-point continuation method for ‘ 1 l regularized minimization with applications to compressed sensing. Technical report CAAM Tech. rep. TR07-07, Rice University 13. Kim S, Koh K, Lustig M, Boyd S, Gorinevsky D (2007) An interior-point method for large-scale ‘1- regularized least squares. IEEE J Sel Top Sig Proc 1(4):606–617

Human Behavior Recognition: An l 1 – l s KSVD-Based Dictionary …

105

14. Kojima M, Megiddo N, Mizuno S (1993) Theoretical convergence of large-step primal-dual interior point algorithms for linear programming. Math Program 59:1–21 15. Malioutov D, Cetin M, Willsky A (2005) Homotopy continuation for sparse signal representation. In: ICASSP 16. Figueiredo M, Nowak R, Wright S (2007) Gradient projection for sparse reconstruction: application to compressed sensing and other inverse problems. IEEE J Sel Top Sig Proc 1(4):586–597 17. Zhang L, Yang M, Feng X (2011) Sparse representation or collaborative representation which helps face recognition?. In: Proceedings of ICCV 18. UCI machine learning repository. Dataset for ADL recognition with wrist-worn accelerometer data set. https://archive.ics.uci.edu/ml/datasets/Dataset+for+ADL+Recognition+with+Wristworn+Accelerometer

Detection and Classification of Breast Cancer in Mammographic Images Using Efficient Image Segmentation Technique Pramit Brata Chanda and Subir Kumar Sarkar

Abstract Breast cancer has become one of the major types of cancer-caused deaths among women of different countries throughout the world. One of the major problems of this type of cancer disease are quick detection or identifying of disease in early stages. In the cases of technologically lagging countries mortality rates are very high due to lack of early diagnosis technology of disease. According to the opinion of different clinical experts, today mammography is one of the most effective diagnosis technologies in medical science domain. So there is a requirement for more accurate methods which can easily diagnose any type of abnormalities in women breast without any kind of human intervention with higher accuracy rates. Segmentation is an approach that is very much required to identify the unambiguous region from the mammogram image. Intensity, texture, and shapes are extracted from the segmented mammogram image. The role of image processing is to detect cancer in human body when input data is in the form of images. For mammogram image classification, the feature extraction of an image with statistical parameter measurement is very important approach. Different types of feature extraction methods are generally used for better classification of abnormality present in mammogram. This technique will provide higher accuracy rates at a comparative higher speed. The statistical parameter includes entropy, mean, regression, correlation, skew, standard deviation. The experimental results achieved 89% accuracy, 74% specificity, and 89% sensitivity, illustrating the usefulness of the technique for identifying and classifying the cancer in mammogram images with maintaining more accuracy. Keywords Breast cancer · Mammogram · Mean · Entropy · Preprocessing · Segmentation · Malignant · Classification · Sensitivity · Morphological P. B. Chanda (B) Computer Science and Engineering, Kalyani Government Engineering College, Kalyani, West Bengal, India e-mail: [email protected] S. K. Sarkar Electronics and Tele Communication Engineering, Jadavpur University, Kolkata, West Bengal, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_9

107

108

P. B. Chanda and S. K. Sarkar

1 Introduction Today, cancer has become huge threat to human life in terms of death over the next few upcoming times. Based on the statistics from the World Health Organization, cancer caused 13 to 40 % of deaths all over the country in 2015 up to 2020 session. Also, due to cancer the rate of deaths gradually increased in the future period, approximately four million deaths can reason from cancer in 2030. The report clearly stated that from different categories of cancer the breast cancer is a type of cancer which can be most frequent type cancer and can be a leading cause of death of females throughout the world. Basically for reducing the death of breast cancer, the screening techniques are very much required. Because the early detection of disease can minimize the risk of critical part of disease, these programs are to provide treatment successfully. Currently, detection of breast cancer is done appropriately with mammography and detection and this has proved to be an effective tool for reduction of the deaths percentage. MIAS Society is a society which is used to provide the mammogram images of cancer. Segmentation includes various approaches for making the mammographical digital image accurate for Artificial Neural Network (Neural Network). The input image consists of different methods which include preprocessing, enhancement of image, noise removal, mass image, targeted image, segmentation, feature extraction. The statistical parameter measurement is an important stage in classification of mammogram. The good features extracted are texture parameter, by which identification of the abnormalities might be found more. Pattern capturing of the image is done by texture-based method. Statistical parameters include skew, entropy, mean, standard deviation, regression, corelation. Classifier used these parameters as input values. Several types of classifiers are used for analysis of image segmentation related applications areas. Here the classification applied on that method using FCM classifier. FCM classifier provides good classification rates and the other factors like better sensitivity, accuracy, etc.

2 Related Previous Work Prof Singh and Sushmita [1] have implemented An Efficient system basis on neural network for identification of breast cancer. Here the author has implemented diagnosis of breast cancer using efficient neural network based classifier. For developing the efficient alternative strategies, they have tested supervised and unsupervised training methods for diagnosis of breast cancer. Back propagation technique has been used by them. Gayathri et al. [2] breast cancer diagnosis using machine Learning algorithms—the work based on survey on detection of breast cancer with various algorithms of machine learning and methods are used for betterment of the accuracy level of cancer prediction. There are huge problems about interpretation of masses in images which are noisy as those images resulting from theacquisition of mammograms [3, 4].

Detection and Classification of Breast Cancer in Mammographic …

109

Furthermore, several research studies undergo and provide indication that more than 70% of breast cancer biopsy surgeries give findings benign [5, 6]. Mammography is still one of the most effective clinical approaches for the early findings of breast cancer [7]. Decision support systems can be important allies for the health professionals for performing diagnostic decision-making more accurately [8]. On the basis of feature extraction strategies, detection and classifying mammary lesions using mammograms is highly dependent, in which the regions of interest, clinically determined by specialist, are preprocessed, and moments, statistics, and other measures are extracted [8]. In breast cancer applications, the use of texture descriptors combined with segmentation methods is very common. However, more complex preprocessing approaches have been used in order to reach higher classification rates by modifying feature dimensionality. One of the most accurate techniques are the series of wavelets: decomposing each image of region of interest by a details images series with different resolutions and an original image of reduced and simplified version of the parts image [9]. These image components are required for extraction of feature. Several methods of this multiresolution approach as for detection of mammary lesions and, in some cases, classify them as benign or malignant findings, with differentiation successfully of the region and the mammary tissue [8, 9]. Mass shape analysis has basically done by Zernike moments [10, 11]. The lesion shape analysis is crucial stages for finding out the degree of malignancy of mammary lesions [12]. Murali and Dinesh [13] have illustrated research work on Classification of Mass in Ultrasound Images using the image segmentation technique. This proposed study is based on ultrasound images that are preprocessed using Gaussian smoothing method for removal of additive noise and using diffusion filters to remove the multiplicative noise (speckle noise). Nithya and Santhi [14] showed the paper on comparative study on feature extraction method for classification of breast cancer. This work represents three different feature extraction techniques for normal and abnormal image classification in mammogram. The work is based on GLCM-based feature extraction method for providing classification rate higher of more than 90%. Khan and Noufal [15] have illustrated, the automatic lesion detection based on wavelet with upgraded active contour method. For this particular work, they have used a method- improved active contour approach that is mainly used in cases of segmentation process for detection of lesion. In this approach, before segmentation wavelet-based and preprocessing of image are done. Here, the mean filter or average filter has been utilized for improving the image quality as per human and the wiener-based filter is used for noise removal. The adaptive median filter has been used as a spatial processing approach for searching pixels of the particular image which are affected by impulse noise [16]. Chethan and Krishna [17] have provided the concepts of breast masses identification in mammograms image using different concentric layers. This method has detected masses automatically in digital mammograms. Array converter converts it to make scaled image. After that binary mask image is generated, thresholding is performed from the grayscale image. The strong connection is done between group of pixels in terms of spatial location and Intensity ranges are illustrated as image granulation.

110

P. B. Chanda and S. K. Sarkar

3 Proposed Methodologies Here, the technique used different steps as like as

3.1 Input Dataset Image In this work, Different Digital Mammogram X-ray Images Dataset (MIAS) are used for experimental case study. We use 19 cancerous images and 22 noncancerous images for detection of breast abnormalities. The biomedical images are always affected by several types of noises that alter the accurate image. The color images (RGB) also can be used for experimental requirements.

3.2 Preprocessing Noises always hamper the performance of biomedical images. Sometimes removal of various categories of noises without changing the desired information becomes a bigger challenge of the research work. Preprocessing of images commonly involves with removal of background noise; each of the images intensity is normalized. Image preprocessing is an approach of enhancing images prior to processing computationally efficient. The preprocessed output is taken from the input data. Preprocessing process use a pixel value of smaller neighborhood for getting fresh brightness value for resultant image. Such type of operation is said as Filtration. The color image is taken as input, so first the image is converted to gray image. This gray image is taken as input to proposed system, then noise parts to be removed. This image converted is into enhanced image, which will be more accurate image to analyze in next step. The first image has been resized to size which is ideal for which the entire work is done. Removal of noise is done using median filter, reducing “salt and pepper” noise is done by it, noise reduction and edges preservation. The direction of input image is taken one side at a time but for other images it will vary particularly from right and left sides.

3.3 Enhanced Image The preprocessed image contains several undesirable white spots or pixels that exists nontargeted region. The contrast limit of image is enhanced or stretched for getting suspected region bright as compared to nontargeted region by giving the pixel threshold value more with higher intensity. The pixels which are unwanted and non-desired areas are removed and rest part to be masked. Then creation of mask is performed.

Detection and Classification of Breast Cancer in Mammographic …

111

Then multiplying reference image and masked image, the target image is created image contains with size of eight bit.

3.4 Morphological Operations Morphological analysis mainly uses step to prior after segmentation. This execution of step is mainly used for tuning the image for going to be segmented or already segmented for executing the accurate clustering methods. Hence, morphological operations are basically used for segmenting properly the images. They are performed for the removal of salt and pepper noise and removal of residual and smoothing of mask (Fig. 1). Input Image(mammogram Cancerous image)

Preprocessing Of Image

Normalized Image

Enhanced Image

Morphological Segmentation

Statistical Analysis with Different Parameters

Image Classification Using FCM Classifier Fig. 1 Proposed methodology

112

P. B. Chanda and S. K. Sarkar

3.5 Segmented Image The process segmentation is used for separating a cancerous image into different segments which pixel sets also called as super pixel. Segmentation is an approach is used to changing the representation of an image more relevant and easier to understande and analyze. The parts are generally separated from background (lines and curves) objects and outsider parts of an image. According to our experimental observation when segmentation methods over, areas of white shaded regions consists the presence of abnormalities in the region as breast area.

4 Results and Analysis According to our experiments some statistical parameter requires for data analysis purposes of the work. These are given in (Table 1). Here, the mean, sd, regression, etc., values are used for statistical measurement. The mean values for benign category tumors are less than the malignant category images. Here the regression values are shown as below 1. It is desirable that the regression coefficient values lies within 1. Here the standard deviation for normal image has higher values than other feature set. Also, execution time for abnormal feature set is shown below. It signifies the values should be as much less as possible. Here the dataset shows the linear classification values with fcm classifier of less mean value and better sd values (Table 2). Table 1 Statistical parameter measurement for proposed segmentation technique Original image

Mean

Standard deviation

Regression

Execution time

Benign (abnormal)

3.2207

17.1669

0.9966

0.8168

Malignant (abnormal)

3.8920

19.6537

0.9974

0.8986

Benign (abnormal)

3.3358

16.6374

0.9964

0.7543

Malignant (abnormal)

3.7943

19.9453

0.9975

0.6891

Normal

5.6535

29.1292

0.9988

1.124

Table 2 Statistical parameter measurement for proposed segmentation technique Original image

Skew

U

Entropy

Benign (abnormal)

3.0497e + 04

2.2617e − 04

5.8075

Malignant (abnormal)

4.5732e + 04

1.5217e − 04

6.4300

Benign (abnormal)

2.5376e + 04

1.9660e − 04

5.6927

Malignant (abnormal)

4.9217e + 04

1.8895e − 04

6.2035

Normal

1.4788e + 05

1.6518e − 04

6.4405

Detection and Classification of Breast Cancer in Mammographic …

113

Here, the statistical analysis shows that the standard deviation value is less for different images. It implies that there is a uniformity exists or deviation or variation of data is less for particular images. In this table for malignant type cancer’s higher standard deviation value than that of benign type cancer affected image. So value of standard deviation is how much lesser that implies that the image is more consistent. Also regression value implies the co-relation between input and the segmented image. Regression parameter value reaches toward to 1 means it becomes more consist anent and accurate for classification. In these particular results the regression parameter are getting value of 0.99 ranges shows the better corelation between images. The mean value is much higher and standard deviation value is lesser for abnormal tumor detected image. Also the method requires lesser execution time for detecting images. Here, the standard deviation values are lesser for malignant and benign type of abnormally classified images than the normal image, which is shown in given picture clearly in the analysis plot (Fig. 2). Here this graph shows the comparative mean, sd values for different normal and abnormal (cancer detected) images. Here the standard deviation values are less for 1st and 3rd images. As per requirement the standard deviation should be lesser for better performance. Here the entropy is higher for 2nd and 4th images and shows better performance. Also, the execution times are very less for all the images. So, the execution time is the crucial factor for better performance. It should be as low as possible (Fig. 3). Here some cancer-affected mammogram images are shown below which are used for early detection of different types of breast cancer. These particular images show the difference between the original and segmented tumor image for comparative measurement (Figs. 4 and 5). These images are used for showing the early detection of malignant tumors after the threshold-based segmentation methods. Here, the early detection of the disease is done properly. The early detection is very important for clinical experts for taking decision about the cases of disease. 35

Parameters

30 25 20 15 10 5 0 Benign (Abnormal)

Malignant Benign Malignant (Abnormal) (Abnormal) (Abnormal) Feature Classes (Tumours) Mean

SD

Fig. 2 Comparison of mean and sd for different tumor images

Normal

114

P. B. Chanda and S. K. Sarkar

Fig. 3 Comparison of entropy and execution time for different tumor images

Fig. 4 a Starting image. b Normalized image. c Segmented benign tumor image

Detection and Classification of Breast Cancer in Mammographic …

115

Fig. 5 a Starting image. b Normalized image. c Segmented malignant tumor image

These values are basically used for feature extraction and classification. There are different feature sets like abnormal, normal, or malignant and benign. So here in case of third image, the accuracy rate is more than 85% to correctly classify the Benign Tumor. Here the experimental performance shows that the sensitivity and accuracy values are higher and False Positive Rate values are less for better classification rates in terms of feature sets. The results show more than 80% rates of classification accuracy as per better confusion parameter’s performance of the classification (Table 3) (Fig. 6). According to the results of different cancerous images taken, it shows the proposed segmentation technique produces better accuracy rates of classifying different features rather than the k-means based segmentation methods. Here, the results proTable 3 Classification accuracies based on feature extraction Images

False positive rate

Sensitivity

Specificity

Accuracy

Benign (abnormal)

0.7648

0.8013

0.7434

0.83846

Malignant (abnormal)

0.7243

0.8138

0.7766

0.8534

Benign (abnormal)

0.7256

0.8922

0.71953

0.8959

Malignant (abnormal)

0.78933

0.8264

0.7378

0.8594

Normal

0.7194

0.8083

0.6907

0.8356

P. B. Chanda and S. K. Sarkar

Accuracy

116

90 88 86 84 82 80 78 76 74 72 Benign

Malignant Tumour (Breast Features)

Kmeans Segmentation

Normal

Proposed Segmentation Methods

Fig. 6 Comparison of classification accuracy for segmentation methods

vide more than 80% of accuracy rates for segmentation and classification of benign and malignant types of cancer diseases.

5 Conclusion Cancer is one of the familiar types for women in several zones of the world. Early detection of these diseases can reduce the probability of death. In these works, a methodology is required for early detection of disease. In our work, the removal of noise is done using different filtering methods that take place using the image enhancement approach that includes the preprocessing of images. Threshold-based segmentation is used for normalizing the image. The morphological operation is used for segmentation of different parts of mammogram. Segmented tumor image ready for statistical measurement of different parameter then confusion matrix parameters are used for measuring classification accuracy. The used algorithm takes higher than 80% accuracy rates for classification, better sensitivity, specificity rates in terms of classification of the different normal and abnormal features of diseases. The results are compared with other research studies for analysing performance that the method performs with better accuracy than k-means based classification approach for classifying the tumor features set. Acknowledgements We use some mammogram images from BRATS datasets and MIAS society images data which are very much important datasets for working this particular domain. So, I acknowledge again my guide and my institute for providing me moral support for this research work.

Detection and Classification of Breast Cancer in Mammographic …

117

References 1. Singh S, Sushmita H (2014) An efficient neural network based system for diagnosis of breast cancer. IJCSIT (BMS Institute of Technology, India) 5(3):4354–4360 2. Gayathri BM, Sumathi CP, Santhanam T (2013) Breast cancer diagnosis using machine learning algorithm—A survay. IJDPS (SDNB Vaishnav College for Women, Chennai, India) 4(3) 3. Rizzi M, D’aloia M, Castagnolo B (2009) Computer aided detection of microcalcifications in digital mammograms adopting a wavelet decomposition. Integr Comput-Aided Eng 16:91–103 4. Rizzi M, D’aloia M, Castagnolo B (2013) A supervised method for microcalcification cluster diagnosis. Integr Comput-Aided Eng 20:157–167 5. Leucht W, Leucht D (2000) Teaching atlas of breast ultrasound. Thieme Medical, New York, pp 24–38 6. Gefen S, Tretiak OJ, Piccoli CW, Donohue KD, Petropulu AP, Shankar PM, Dumane VA, Huang L, Kutay MA, Genis V, Forsberg F, Reid JM, Goldberg BB (2003) ROC analysis of ultrasound tissue characterization classifiers for breast cancer diagnosis. IEEE Trans Med Imaging 22(2):170–177 7. Prognostic Factors in Breast Cancer, according Brazilian Government (in Portuguese). http:// www.inca.gov.br/rbc/n_48/v01/pdf/revisao.pdf. Accessed Jan 2011 8. de Lima SM, da Silva-Filho AG, dos Santos WP (2013) Detection and classification of masses in mammographic images in a multi-kernel approach. Comput Methods Programs Biomed. http://dx.doi.org/10.1016/j.rgo.2013.10.012 9. Nascimento MZ, Martins AS, Neves LA, Ramos RP, Flores LE, Carrijo GA (2013) Classification of masses in mammographic image using wavelet domain features and polynomial classifier. Expert Syst Appl 40(1):6213–6221 10. Tahmasbi A, Saki F, Shokouhi SB (2011) Classification of benign and malignant masses based on zernike moments. Comput Biol Med 41(1):726–735 11. Saki F, Tahmasbi A, Soltanian-Zadeh H, Shokouhi SB (2013) Fast opposite weight learning rules with application in breast cancer diagnosis. Comput Biol Med 43(1):32–41 12. BI-RADSTM (Breast Imaging Reporting and Data System) (2003) American college of radiology, Fourth edn 13. Murali S, Dinesh MS (2012) Classification of mass in breast ultrasound images using image processing techniques. IJOCA (Mysore University, India) 42(10) 14. Nithya R, Santhi B (2005–2011) Comparative study on feature extraction method for breast cancer classification. JATIT & LLS (School of Computing, SASTRA University) 33(2) 15. Khan AK, Noufal P (2014) Wavelet based automatic lesion detection using improved active contour method. IJERT Dep Electron Commun (MES College of Engineering, Kuttippuram, Malappurum, Kerala) 3(6) 16. Naranje S (2016) Early detection of breast cancer using ann. IJIRSET 4(7) 17. Jai-Andaloussi S, Sekkaki A, Quellec G, Lamard M, Cazuguel G, Roux C (2013) Mass segmentation in mammograms by using bidimensional empirical mode decomposition BEMD. In: 35th annual international conference of the IEEE EMB, Osaka, Japan, 3–7 July 2013

Energy Systems

Visualization and Improvement of Voltage Stability Region Using P-Q Curve Srijan Seal and Debjani Bhattacharya

Abstract Voltage instability in power system is becoming more and more important because of the regular growth of power system and lack of efficiency in reactive power management. The voltage instability of a power system is associated with a voltage drop. Voltage drop has a cumulative effect unless efficient reactive power sources are available for voltage regulation. In this paper, the voltage stability region of IEEE standard bus systems is visualized using P-Q curve technique. Voltage stability margins are also visualized using local measurement techniques. Then, the change in stability margin is observed by introducing FACTS and DG. Optimum location of installation of these devices is determined. Keywords Voltage stability · Thevenin’s equivalent · P-Q curve · FACTS · Distributed generation · Local measurement

1 Introduction In recent years, voltage instability or voltage collapse have caused large-scale blackouts throughout the world including Scandinavia (2003), Northeastern United States (2003), Athens (2004), Brazil (2009) and India (2012). So, visualization of voltage stability region and its improvement is of utmost importance. Keeping in view the growing consumer demands, Smart Grid technology and uses of small-scale Distributed Generations, the power system has become one of the most complex engineering systems. So it is important to analyse voltage collapse possibility and areas of improvement of the voltage profile. With the development of excitation systems, faster short circuit clearing time and other transient state stability control devices, the problems due to transient state S. Seal · D. Bhattacharya (B) Department of Electrical Engineering, Academy of Technology, Adisaptagram, Hooghly 712121, India e-mail: [email protected] S. Seal e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_10

121

122

S. Seal and D. Bhattacharya

stability is now largely reduced. But the problems regarding steady-state stability cannot be easily dealt with because of complicated power system network, variable and huge load demands and generating and transmitting constraints. So steady-state stability analysis is a major concern. Voltage is considered to be one of the most useful system responses in case of steady-state stability analysis. A system is said to be in voltage-stable condition under a certain operating condition, if it can maintain its original state after being subjected to a small disturbance. The stability mainly depends on the reactive power present in the system. Proper balance of lagging reactive power generation and load demand can increase the system stability. There are many methods to predict the possibility of voltage collapse or the system instability, some of those are: Online monitoring of voltage stability margin [1], P-V, Q-V curve analysis technique [2], Modal Stability analysis [3], Minimum singular value of Jacobian matrix of N-R load flow analysis [4], Global security indicator for an equivalent power network [5]. In this paper, an attempt has been made to find out the weakest bus in standard IEEE 14 Bus, 30 Bus and 57 Bus test systems using P-Q curve analysis technique. The P-Q curves are plotted. The area under P-Q curve is considered to be the stability index. The result is compared with other traditional methods of stability analysis. It reveals that it is possible to use this method to identify the weakest bus of a power system network. An attempt has been made to enhance the system stability by introducing Distributed Generation and FACTS devices. The changes in stability indices are observed after installation of the above. Optimum location of installation of the DG and FACTS is determined using simulation.

2 Methodology A. Thevenin’s Equivalent model of power system According to Thevenin’s theorem, we can represent each bus of a given network as a simple two-bus model as shown in Fig. 1. Here, V T = Thevenin’s Equivalent voltage of the bus, Z T = Thevenin Equivalent impedance of the bus, Z L = the series representation of connected load to the bus. Here Z L is represented as series resistance (RL ) and series reactance (XL ) using the Fig. 1 Thevenin’s equivalent of a load bus

Visualization and Improvement of Voltage Stability Region …

123

following equations: R L p.u. = VL2p.u Sb X L p.u. = VL2p.u Sb

Q L p.u. PL2 p.u + Q 2L p.u Q L p.u. PL2 p.u

+ Q 2L p.u

(1) (2)

where, VL p.u = load voltage and S b = Base MVA. ZBUS matrix is constructed for determining ZTH . The diagonal element (Zkk ) of ZBUS matrix actually represents the ZTH of the system corresponding to the kth bus. To find VTH load flow analysis is run. If VL be the voltage at a load bus then VTH can be computed using Eq. (3). VT H = (1 + Z T H /Z L )VL

(3)

B. Online Monitoring of Voltage Stability Margin This method calculates a stability index based on basic definition of voltage stability, using local measurements. It is very simple and straight forward and is proven to be computationally efficient for online monitoring of voltage stability. The steps involved in this method are as follows: From Fig. 1 the magnitude of the current through load can be given by VTH IL =  (ZTH cos θ + ZL cos φ)2 + (ZTH sin θ + ZL sin φ)2 VTH Or, I L =  Z2TH + Z2L + 2ZTH ZL cos(θ − φ)

(4)

Here, θ phase angle of impedance ZTH and φ phase angle of impedance ZL The magnitude of the receiving end voltage is vTH zL VR = ZL I L =  z2TH + Z2L + 2ZTH zL cos(θ − φ)

(5)

The apparent power supplied to the load is ∴S=

V2TH ZL Z2TH + Z2L + 2ZTH ZL cos(θ − φ)

At maximum load apparent power,

dS dY

=0

(6)

124

S. Seal and D. Bhattacharya

  v2TH 1 − Y2 z2TH dS = =0 dY (1 + z2TH Y2 + 2ZTH Y cos(θ − φ))2

(7)

Solution of the above equation gives, ZTH = ZL . The above equation indicates, ZTH is the maximum loading point. Hence, in the dS versus Y, the maximum load admittance can be identified as the point plot of dY dS where dY = 0. Obviously, the weakest bus will have the lowest value of maximum load admittance. C. Stability region visualization using P–Q curve This method is based on a very basic idea that there can be no feasible stability points outside the limiting values of active and reactive power in the P-Q plane. This method depicts the voltage stability limit graphically and the calculations involved are faster than most of the other methods. Hence it is both computationally efficient and easy to visualize. From Fig. 1, the load current IL can be represented as   I L = PL − jQL /VL∗

(8)

V L = VTH − ZTH I L

(9)

And,

Using Eqs. (8) and (9), the load voltage VL can be obtained as   V4L + 2(RTH PL + XTH QL )VL2 − VL2 V2TH + R2TH + X2TH (P2L + Q2L ) = 0

(10)

At the verge of voltage collapse, the discriminant of the above equation becomes zero, and hence the following equation is given as   V4TH + 4 2PL QL RTH XTH − V2TH (RTH PL + XTH QL ) − R2TH Q2L − X2TH P2L = 0 (11) Or, f(PL , QL , VTH , RTH , XTH ) = 0

(12)

From the above equation, considering only the positive value of QL , we get 

1 −4R3TH V2TH PL + R2TH V4TH + 15R2TH X2TH P2L − 16RTH V2TH X2TH PL + 4V4TH X2TH R2TH − 2V2TH XTH + 4RTH XTH PL ) (13)

QL =

Hence, by varying PL up to certain limit, the P-Q plot can be obtained. A typical such plot is given in Fig. 2. Since no feasible load points can be outside the given curve the area under the P-Q curve can be used as an index to determine voltage stability region of the bus.

Visualization and Improvement of Voltage Stability Region …

125

Fig. 2 Typical P-Q curve

The Trapezoidal method is used to determine the area under the curve and the area is designat as P-Q index. The bus having the lowest area, i.e., lowest P-Q index is considered to be the weakest bus. D. Improvement of Voltage Stability Region and modeling of FACTS and DG The most effective way to enhance system stability is to provide installation of FACTS or Distributed Generation (DG). Installation of these, the real and reactive power injection of the bus are changed, which in turn changes VTH . Here the FACTs are considered as the reactive power sources and DG’s are considered as real as well as reactive power sources. DG units based on synchronous machine for small hydro, geothermal, and combined cycles may be considered as DG generating both active power and reactive power. Fuel cells, photovoltaic cells, micro turbines integrated to the main grid can be considered as sources of active power only. The DG units equipped with synchronous compensator may be considered as generating reactive power only.

3 Simulation and Result A. Identification of the weakest bus The P-Q curve technique is implemented on an IEEE 14 bus, 30 bus and 57 bus test system and the results are compared with the online monitoring technique. Both the methods of identifying weakest bus give same result. The results obtained are given in Table 1. From Table 1 it can be seen that the P-Q index i.e. the area under the P-Q curve is minimum for bus 12 in IEEE 14 bus test system, for bus 26 in IEEE 30 bus test system and for bus 31 in IEEE 57 bus system. It can also be seen from the table that the value of load admittance Y for which dS/dY is zero is minimum for

126

S. Seal and D. Bhattacharya

Table 1 Identification of weakest bus IEEE 14 bus system

IEEE 30 bus system

IEEE 57 bus system

Bus No.

Y at dS/dY =0

P-Q index

Bus No.

Y at dS/dY =0

P-Q index

Bus No.

Y at dS/dY =0

P-Q index

11

1.91

0.359

26

1

0.081

31

0.95

0.183

12

1.66

0.287

29

1.19

0.107

32

1.02

0.222

14

1.81

0.365

30

1.13

0.116

33

0.97

0.198

bus 12, bus 26 and bus 31 in IEEE 14 bus, 30 bus and 57 bus system respectively. The graphical representations are shown from Figs. 3, 4, 5, 6, 7 and 8. From Figs. 6, 7 and 8 buses having minimum area under P-Q curve can be easily identified. Fig. 3 Variation of dS/dY with change in load admittance for different buses of IEEE 14 bus system

Fig. 4 Variation of dS/dY with change in load admittance for different buses of IEEE 30 bus system

Visualization and Improvement of Voltage Stability Region …

127

Fig. 5 Variation of dS/dY with change in load admittance for different buses of IEEE 57 bus system

Fig. 6 P-Q curve for different buses of IEEE 14 bus system

Fig. 7 P-Q curve for different buses of IEEE 30 bus system

B. Improvement of voltage stability region For the improvement of voltage stability region DG units of fixed capacity (10 MW active and 10 MVAR reactive power) are installed at the weakest bus. The analysis is performed for four cases. Case I: the base case without DG and FACTS, Case II: considering FACTS, Case III: considering DG units generating real power only

128

S. Seal and D. Bhattacharya

Fig. 8 P-Q curve for different buses of IEEE 57 bus system

and Case IV: considering DG units generating both real and reactive power. All the results obtained are compared with Case I. The results obtained are given in Table 2 and graphical representations are shown in Figs. 9, 10 and 11. Table 2 Improvement of voltage stability region Bus system

Weakest bus No.

P-Q index Base case without any external device

With DG generating P only

With Facts

With DG generating both P and Q

IEEE 14 bus system

12

0.287

0.296

0.3

0.309

IEEE 30 bus system

26

0.081

0.092

0.101

0.113

IEEE 57 bus system

31

0.183

0.194

0.224

0.236

Fig. 9 Improvement of P-Q index of IEEE 14 bus system

Visualization and Improvement of Voltage Stability Region …

129

Fig. 10 Improvement of P-Q index of IEEE 30 bus system

Fig. 11 Improvement of P-Q index of IEEE 57 bus system

C. Optimum location of DG and FACTS Optimal location of DG unit is obtained by placing DG unit of same size at other buses. The P-Q index obtained is tabulated in Tables 3, 4 and 5. From the tables it is clear that when the DG is placed at the weakest bus the P-Q index of that bus is maximum. So that bus is the optimum location of the DG. The same analysis is done to find the optimal location of FACTS (Fig. 12, 13, 14).

130

S. Seal and D. Bhattacharya

Table 3 Optimal location of facts and DG units for IEEE 14 bus system Bus No.

P-Q index With DG generating P Only

With FACTS generating Q only

With DG generating Both P and Q

4

0.287

0.286

0.286

5

0.287

0.286

0.286

7

0.287

0.286

0.286

9

0.286

0.286

0.286

10

0.286

0.286

0.286

11

0.287

0.286

0.286

12

0.296

0.3

0.309

13

0.288

0.291

0.292

14

0.287

0.288

0.288

Table 4 Optimal location of facts and DG units for IEEE 30 bus system Bus No.

P-Q index With DG generating P Only

With FACTS generating Q only

With DG generating both P and Q

3

0.081

0.081

0.081

4

0.081

0.081

0.081

6

0.081

0.081

0.082

7

0.081

0.081

0.081

9

0.081

0.082

0.082

10

0.081

0.082

0.083

12

0.081

0.082

0.081

14

0.081

0.082

0.082

15

0.081

0.082

0.082

16

0.081

0.082

0.082

17

0.081

0.082

0.082

18

0.081

0.082

0.082

19

0.081

0.082

0.082

20

0.081

0.082

0.082

21

0.081

0.083

0.083

22

0.082

0.083

0.084

23

0.081

0.083

0.083

24

0.082

0.085

0.086

25

0.084

0.09

0.094 (continued)

Visualization and Improvement of Voltage Stability Region …

131

Table 4 (continued) Bus No.

P-Q index With DG generating P Only

With FACTS generating Q only

With DG generating both P and Q

26

0.092

0.1

0.11

27

0.082

0.087

0.089

28

0.081

0.082

0.082

29

0.082

0.087

0.088

30

0.083

0.087

0.088

Table 5 Optimal location of facts and DG units for IEEE 57 bus system

Bus No.

P-Q index DG generating P only

FACTS generating Q only

DG generating both P and Q

4

0.183

0.183

0.183

5

0.183

0.183

0.183

7

0.183

0.184

0.184

10

0.183

0.184

0.183

11

0.183

0.184

0.184

13

0.183

0.184

0.184

14

0.183

0.184

0.184

15

0.183

0.184

0.184

16

0.183

0.183

0.183

17

0.183

0.183

0.183

18

0.183

0.184

0.184

19

0.184

0.185

0.185

20

0.184

0.185

0.186

21

0.184

0.186

0.187

22

0.184

0.187

0.188

23

0.184

0.187

0.188

24

0.186

0.19

0.194

25

0.184

0.206

0.207

26

0.186

0.19

0.193

27

0.185

0.187

0.188

28

0.184

0.185

0.186

29

0.183

0.185

0.185 (continued)

132 Table 5 (continued)

Fig. 12 Optimal location of FACTS and DG units for IEEE 14 bus system

S. Seal and D. Bhattacharya Bus No.

P-Q index DG generating P only

FACTS generating Q only

DG generating both P and Q

30

0.187

0.211

0.215

31

0.194

0.224

0.236

32

0.185

0.209

0.211

33

0.185

0.209

0.211

34

0.187

0.191

0.195

35

0.186

0.19

0.192

36

0.185

0.189

0.191

37

0.185

0.188

0.190

38

0.184

0.186

0.187

39

0.185

0.188

0.189

40

0.185

0.188

0.19

41

0.183

0.185

0.185

42

0.184

0.185

0.186

43

0.183

0.184

0.184

44

0.184

0.186

0.186

45

0.183

0.185

0.185

46

0.183

0.185

0.185

47

0.183

0.185

0.186

48

0.184

0.186

0.186

49

0.183

0.185

0.185

50

0.183

0.185

0.185

Visualization and Improvement of Voltage Stability Region …

133

Fig. 13 Optimal location of FACTS and DG units for IEEE 30 bus system

Fig. 14 Optimal location of FACTS and DG units for IEEE 57 bus system

4 Conclusion In this paper, the P-Q index has been used as a voltage stability index. This P-Q index has been used to identify the weakest bus in the IEEE 14, 30, 57 bus test system. The P-Q index of the buses is determined using Thevenin’s theorem. This approach avoids repetitive load flow analysis for determining the P-Q curves. Effects of FACTS devices and DG units on P-Q index are tabulated and also represented graphically. It has been seen that in all these cases the P-Q index increases from the base case where no FACTS and DG are installed. Finally, the optimal location of FACTS and DG unit has been obtained where the voltage stability (i.e., the P-Q index) is maximum.

References 1. Phadke AR, Fozdar M, Niazi KR (2008) A new technique for online monitoring of voltage stability margin using local signals. In: Fifteenth national power system conference (NPSC), IIT

134

S. Seal and D. Bhattacharya

Bombay 2. Sharma P, Kumar A (2016) Thevenin’s equivalent based P-Q-V voltage stability region visualization and enhancement with FACTS and HVDC. Int J Electr Power Energy Syst 80:119–127 3. Gao B, Morison GK, Kundur P (1992) Voltage stability evaluation using modal analysis. Trans Power Syst 7(4) 4. Chakraborty K, Biswas SD (2007) An offline simulation method to identify the weakest bus and its voltage stability margin in a multibus power network. In: Proceedings of international conference MS’07 5. Dey S, Chanda CK, Chakrabarti A (2004) Development of a global voltage security indicator and role of SVC on it in LPS system. J Electr Power Syst Res (USA) 68

Analysis of Temperature at Substrate and Sink Area of 5 W COB-Type LEDs, with and Without Driver Debashis Raul

Abstract In recent year, Light-Emitting Diodes (LEDs) have been widely used due to their excellent advantages over conventional light sources, e.g., like incandescent lamps, gas discharge lamps, etc., with their high efficacy, low power consumption, and long lifetime. Reliability and long lifetime will determine the economical and ecological success of LED in lighting systems. LED systems can in general reach very long lifetime of up to 200,000 h with appropriate designed. For solid state, LEDs, the light output, efficiency, spectral distribution and lifetime is strongly dependent on operating temperature. In this study, the thermal tests under ambient temperature have been performed. The temperature build up at the substrate area and sink area has been measured by the Thermal IR imager at every 5 min interval. Two processes have been adopted. Initially the measurement of temperature was done and recorded when the LEDs are connected with constant 300 mA driver and another is when LEDs are connected with rated DC power supply, i.e., without driver condition. To find the extra heat generation using driver or not. But the temperature generation is nearly same for both the conditions, however details experimental results has been furnished and analyzed herein after. Keywords Chip-on-board (COB) · Correlated color temperature (CCT) · Substrate area · Sink area

1 Introduction In present world, artificial lighting system plays a major role for development. Since illumination systems consume about 20% of total electrical power consumption of India [1]. Similar picture is also for other countries. It calls for development of energyefficient light sources. Solid state lighting system, i.e., Light-Emitting Diodes (LEDs) are a most energy-efficient prospective light sources compare to other conventional lighting sources. LED lamps have a lifetime and electrical efficiency which are several D. Raul (B) School of Illumination Science Engineering & Design, Jadavpur University, Kolkata 700032, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_11

135

136

D. Raul

times greater than incandescent lamps and other gas discharge lamps. Some LED chips are able to emit more than 200 lm/W whereas for incandescent lamps the efficacies are 5–15 lm/W. A maximum portion of the electricity in an LED converts as heat rather than light (about 75% heat and 25% light). LED lamps are badly affected by high temperature, which reduced the light output and life of these. For this purpose, LED lamps are required proper thermal management to decrease the chip temperature. So LED lamps typically include heat dissipation elements such as heat sinks and various cooling systems are required [2–4]. Elger et al. [5] developed a method that measured the relative thermal resistance of the LED based on transient thermal analysis and using this method they predicted the accurate lifetime of it. Since LED is an electronic device if this heat is not removed, the LEDs would run at high temperature, which not only lowers their efficacy, but also makes the LED less reliable. Chen et al. [6] evaluated the reliability of the LED packages under thermal cycling and thermal shock conditions and monitored the optical degradation and electrical parameters variation of the LED package. Chen et al. [7] carried out an experiment inside the LED package by an online testing method to find the LEDs’ optical degradation under with and without cooling system. When the forward biased is applied, the temperature of the junction of the LED is generated which is called junction temperature of it. So it is needed to minimize the junction temperature to maintain the desired LED lifetime and performance [8, 9]. Chen et al. [10] developed a portable junction temperature instrument which measured the dynamic junction temperature which is most important parameter of the LED. Tang et al. [11] experimented by using three different Thermal Interface Materials (TIM) like Silver paste, Tin alloy and grapheme of the LED. After that they investigated the thermal behavior of the LED and found the grapheme is the best TIM material for heat dissipation from LED chip to Sink. From field studies it has been observed that the declared performance by manufacturer about LED never tallies with that of field. This calls for thorough study on the subject for lamp as well as system design optimization of LED-based system. In this experiment the temperature measurement at substrate and sink area of the COB-type LEDs have been carried out at ambient condition, i.e., at room temperature. These values of temperature rise at those areas are essential in lamp as well as system design criteria.

2 Experiment 2.1 Test Samples In this experiment two groups of commercially available LED lamps are used and each group has five lamps. All the LEDs are of same make. In Group#: 1, 5 W COB-type warm white LEDs are used and in Group#: 2, 5 W COB-type cool white

Analysis of Temperature at Substrate and Sink Area of 5 W …

137

Table 1 Specification of LEDs Type of LED

No. of LED

Power rating (W)

Dimensionsa (mm) D()

H

CCT (K)

Material of heat sink

Electric circuit of LED matrices

COB warm white

5

5

88

40

3000

Aluminum alloy

Compact in-built chip

COB cool white

5

5

88

40

6500

Aluminum alloy

Compact in-built chip

a D()—Diameter,

H—Height

LEDs are used. The CCT of the five COB-type LEDs are 3000 K (Warm White) and another five are 6500 K (Cool White) Details of the LEDs are shown in Table 1.

2.2 Experimental Procedure An experiment had been conducted using two groups of COB-type LED lamps. The temperatures at substrate and sink area of the LEDs are measured when lamps are connected by the constant 300 mA driver circuit. The temperatures at various parts are of the LEDs measured by using a Thermal Imager which is shown in Fig. 1. The

Fig. 1 Set up for measurement of temperature of LED assembly by using thermal imager (Fluke make Ti 400)

138

D. Raul

Fig. 2 Set up for measurement of LEDs’ current and illuminance value

temperatures at substrate area are measured by 5 min time interval up to one hour. Then temperatures have been measured at sink area for all the LEDs at the same time interval. After that, the same LEDs are used to measure the temperature at substrate and sink area with using the rated DC power supply, i.e., without driver condition. At the same time, the illuminance values have been measured by the chromameter of the LEDs. The chromameter was placed on the table at a distance of two feet apart from the LED luminaires. Correlated color temperature (CCT) also measured with the help of chromameter (CL 200A, Konica Milonta make). Now the driver is connected to the AC mains and measured the output voltage, i.e., the DC rated output voltage by Multimeter. Now the measured DC voltage are 14.45 and 16.3 V for COB Warm and Cool white LED respectively. Now provide the DC supply to the LEDs and measured the diode current as well as the illuminance values simultaneously, as shown in Fig. 2. For both the LEDs the ambient temperature was at 29 °C when the measurements have been taken. All the above measurements were repeated of five times for all the methods and mean values are used all graphical plots.

3 Results and Discussions Two groups of the same-make LEDs were used in the experiment. Each group had a population of five. Now the temperatures were measured of substrate and sink area of all the LEDs and plot the data from each group with time. Now for COB-type warm white LED the temperature variation is not linear, this is shown in Fig. 3a. Because the generation of the temperature of the junction of this LED is dissipated toward the backside of the chip to heat sink area of it and the dissipated heat exit to the environment. Here the temperature data is measured at an interval of 5 min.

Analysis of Temperature at Substrate and Sink Area of 5 W …

139

Fig. 3 Graph for temperature variation with time at the a substrate area, b sink area of COB-type 5 W warm White LED with using 300 mA driver

At that time the LED produced the heat that indicates the temperature peak in the below curve. The temperature variation at sink area is shown in Fig. 3b. Where, the temperatures of both type LEDs are increased for first 15 min of their burning time. After that the temperature of the substrate area observed as decrease for 5 min and then the temperature again increased. In this way the temperatures are varied with respect to their burning time till steady state remains and equilibrium is reached through the total generated heat distribution by conduction. The temperature characteristics curve for COB-type 5 W cool white LED at substrate area and sink area are shown in the Fig. 4a, b respectively. Those curves show that the temperature variation at the substrate area and the sink area has almost same. The temperature is raised up to 63.4 °C for first 10 min of its burning time for the COB-type cool white LED. After 10 min for its burning time to one hour the average temperature of the LED is 64.2 °C. For both the LEDs the ambient temperature was at 29 °C when the measurements have been taken.

140

D. Raul

Fig. 4 Graph for temperature variation with time at the a substrate area, b sink area of COB-type 5 W Cool White LED with using 300 mA driver

Now, the measured data of temperature at the sink area for COB 5 W Warm White LED with driver is compared with and without driver condition which is shown in Fig. 5a. From this characteristics curve demonstrates that the temperature variation of COB 5 W warm white LED is looks same but the difference is in its magnitude of temperature generation of this LED. More temperature is generated by this LED when it is run with DC power supply (without driver condition). In Fig. 5b, for COBtype cool white LED the temperature variation for both the condition is little bit same but the produced temperatures are nearly same. A LED is a p–n junction device. When it is applied the forward voltage then it will pass diode current. Now the forward voltage and forward current are in an exponential relation. Here the measured value of the diode current and applied forward voltage gives the LED’s characteristics curve which is shown in Fig. 6. The measured results can be used to check if there is any electric or material change of the LEDs.

Analysis of Temperature at Substrate and Sink Area of 5 W …

141

Fig. 5 Graph for temperature variation with time at the sink area of COB 5 W. a Cool, b warm White LED with and without driver

The following Fig. 6a, b show the test result of one sample from group 1 and group 2 respectively. Now it is concluded that electrical properties of the LEDs were very stable during the entire experiment. In this experiment, the Correlated Color Temperatures (CCT) were measured for both types of LEDs to find any color shift. But the test results indicate that the CCT of the LEDs are nearly stable which is shown in Table 2 of one sample from each Group. The CCT values are represents by the CIE 1976 Chromaticity diagram for both type of LEDs in Fig. 7. If the CCT value of LED shifts from its quadrangle, then the LED will be concluded as failed in stability of color.

142

D. Raul

Fig. 6 a Voltage-current characteristics of COB-type warm white 5 W LED. b Voltage-current characteristics of COB-type cool white 5 W LED Table 2 Correlated color temperature for COB-type LEDs

Time (min)

COB warm white LED

COB cool white LED

Initial

2986

5126

1

2990

5283

5

2996

5328

10

3008

5238

15

3021

5248

20

3023

5210

25

3012

5171

30

2950

5171 (continued)

Analysis of Temperature at Substrate and Sink Area of 5 W … Table 2 (continued)

143

Time (min)

COB warm white LED

COB cool white LED

35

3024

5171

40

3026

5169

45

3022

5169

50

3021

5164

55

3021

5171

60

3020

5171

Fig. 7 CIE 1976-Chromaticity diagram for COB 5 W. a Warm and b Cool white LED

144

D. Raul

4 Conclusion Several experiments have been conducted for analyzing nature of light output, thermal characteristics of the same make of COB-type LEDs. In present days COB-type LEDs have wide application in lighting field. Performance analyses on COB type of LEDs are required. Since, it is a consumer satisfaction in the commercial aspect so light output nature in different conditions should need to be known to all, especially for manufacturers as well as designers. By this analysis designer can modify the product as per need of application. LEDs are switched on and measured the temperature for a fixed time interval at the substrate area and sink area which is analyzed. If it might have defects on the driver so observations have been made by DC power by providing to LEDs as inputs and temperatures at the sink area have been measured. Then comparisons between the characteristics curves of with driver and without driver have been made to show the results of the LEDs are almost same. It has been observed that the measured data of temperature at the sink area for COB 5 W Warm White LED with driver and without the temperature variation of COB 5 W warm white LED is looks same but the difference is in its magnitude of temperature generation of this LED. More temperature is generated by this LED when it is run with DC power supply (without driver condition). For COB-type cool white LED the temperature variation for both the condition is little bit same but the produced temperatures are nearly same. Acknowledgements The authors wish to acknowledge the Government of West Bengal, India for providing the fellowship. Authors wish to acknowledge the support received from School of Illumination Sciences, Engineering & Design of Jadavpur University for facilitate with experimental set up to complete this work at the laboratory, therein.

References 1. http://ujala.gov.in/documents/Final_Monitoring_and_Verificatio_Report_EESL.pdf 2. Zhong D, Zhang J (2011) Thermal and optical simulation of high-power LED array based on Silicon heatsink. IEEE, 978-1-4577-0321-8/11 3. Kiseev V, Aminev D, Sazhin O (2017) Two-phase nanofluid-based thermal management systems for LED cooling. STPM2017 IOP Conference on series: materials science and engineering. https://doi.org/10.1088/1757-899x/192/1/012020 4. Mohamad MS, Abdullah MZ, Abdullah MK (2013) Experimental study on the cooling performance of high power LED arrays under natural convection. In: 2nd international conference on mechanical engineering research (ICMER 2013) IOP conference on series: materials science and engineering. https://doi.org/10.1088/1757-899x/50/1/012030 5. Elger G, Hanss A, Schmid M, Wipiejewski T (2014) Application of thermal analysis for the development of reliable high power LED modules. IEEE, 978-1-4799-6697-4 6. Chen Z, Zhang Q, Chen R, Jiao F, Chen M, Luo X, Liu S (2011) Comparison of LED package reliability under thermal cycling and thermal shock conditions by experimental testing and finite element simulation. In: Electronic components and technology conference (ECTC). IEEE. https://doi.org/10.1109/ectc.2011.5898550

Analysis of Temperature at Substrate and Sink Area of 5 W …

145

7. Chen Q, Chen Q, Hu R, Luo X (2014) Is thermal management outside the package enough for higher LED reliability. IEEE, 978-1-4799-4707-2/14 8. Chhajed S, Xi Y, Gessmanna T, Xi J-Q, Shah JM, Kim JK, Schubert EF (2005) Junction temperature in light-emitting diodes assessed by different methods. Rensselaer Polytechnic Institute Troy, NY, p 12180 9. Narendran N, Liu Y-W, LED life versus LED system life. Lighting Research Center Rensselaer Polytechnic Institute Troy, NY 10. Chen Q, Luo X, Zhou S, Liu S (2011) Dynamic junction temperature measurement for high power light emitting diodes. American Institute of Physics, 0034-6748/2011/82(8)/084904/7 11. Tang Y, Liu D, Yang H, Yang P (2016) Thermal effects on LED lamp with different thermal interface materials. IEEE Trans Electron Devices 63(12):4819–4824. https://doi.org/10.1109/ TED.2016.2615882

Performance Study and Stability Analysis of an LED Driver Piyali Ganguly, Vishwanath Gupta and Parthasarathi Satvaya

Abstract The present work deals with performance study and stability analysis of an LED driver system. An LED driver based on buck-boost topology is designed and simulated in MATLAB Simulink environment. The driver satisfactorily operates LED modules having power rating in the range of 6 W to 24 W. The power factor and Total Harmonic Distortion comply with standard recommended values. The mathematical model of the LED driver is formulated and the stability analysis of the designed driver is carried out during its operation. Keywords LED driver · Design and simulation · Stability analysis · Electrical performance

1 Introduction Lighting is responsible for 20% of the total energy consumption, therefore recent trends is to develop more energy efficient lighting with high efficacy. Solid-state lighting is presently used as an efficient lighting source for commercial and residential lighting applications as it has longer lifetime and the absence of toxic mercury content compared with other light sources [1]. LED lighting system, when designed correctly, can last for more than 100,000 operating hours. LED lighting applications operated on AC should employ drivers with high energy efficiency and for variation of output power it should employ a dc–dc converter and various control strategies. Various authors use different DC–DC converter techniques for designing LED drivers. In [2] Jha and Singh proposed a bridgeless ZETA converter used in low-voltage multistage LED. In [3] Sulistiyanto et al. proposed an LED driver design based on synchronous P. Ganguly (B) Department of Electrical Engineering, Seacom Engineering College, Howrah, India e-mail: [email protected] V. Gupta Electrical Engineering Department, Jadavpur University, Kolkata, India P. Satvaya School of Illumination Science Engineering and Design, Jadavpur University, Kolkata, India © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_12

147

148

P. Ganguly et al.

buck power converter. In [4] Sexena and Kulshrestha used a single-stage PFC fourthorder dc–dc buck converter having inductor at source side to design the proposed LED driver. However, in the above-mentioned work, the LED driver was incapable of operating LED modules of different power ratings and their stability analysis was not discussed. The objectives of the present work are to model, simulate, and obtain its electrical performance when it operates 6–24 W LED modules at rated AC supply. The steadystate stability analysis and the designed LED driver is also carried out.

2 Electrical Parameters of LED Modules For the proper design of a LED driver, the LED load has to be modeled. The linear model of 6–24 W assembled with 1.2 W chips is done using the MATLAB SIMULINK model which consists of threshold voltage and dynamic resistance in series as given by Eq. 1 [5, 6]. VD = RDi IDi + Vγi

(1)

where VD , RDi , VGi , and IDi are the rated voltage, dynamic resistance, threshold voltage and current of single LED chip respectively. If we connect the LED chips connected in series–parallel combination, the expression for output voltage becomes as given by (2)  Vo =

 NS (RDi ) ∗ Np IDi + Ns VYi = RD Io + Vγ Np

(2)

where Vo , RD , Io , and VG are the rated voltage, dynamic resistance, rated current and threshold voltage of the entire LED module respectively; NS and NP are the number of LED chips connected in series and parallel respectively. Now for this modeling process a single LED chip of 1.2 W was tested in the laboratory. The laboratory set up for this experiment is shown in Fig. 1. The current values for several forward voltages had been gathered in Table 1 and by using these values a best-fit curve was drawn in MATLAB environment as shown in Fig. 2. From this characteristic curve the value of VG , and RD are obtained as Vγ = 8.3V RD = 36 Next, LED load of different wattages have been designed according to different series and parallel combination of LED chips. The electrical parameters of LED modules from 6 to 24 W are given in Table 2.

Performance Study and Stability Analysis of an LED Driver

149

Fig. 1 Experimental setup for testing a single LED chip of 1.2 W

Table 1 V-I characteristics of test LED chip

Test forward voltage (V)

Test forward current (A)

RD ()

VG (V)

8

0.001

36

8.3

8.3

0.005

8.6

0.01

8.9

0.016

9.2

0.023

9.5

0.03

9.8

0.038

10.1

0.046

10.4

0.055

10.7

0.064

11.0

0.074

11.3

0.084

11.6

0.094

11.9

0.106

12.0

0.11

Fig. 2 Best fit curve for I-V characteristics of single LED chip

150

P. Ganguly et al.

Table 2 Electrical parameters of proposed LED modules Series–parallel Load (W) combinat on of LED Module (NS × NP )

Rated output voltage (V)

Rated output current (A)

Dynamic resistance ()

Threshold voltage (V)

4×2

9.6

48

0.204

72

33.2

4×3

14.4

48

0.306

48

33.2

4×4

19.2

48

0.408

36

33.2

5×1

6

60

0.102

180

41.5

5×2

12

60

0.204

90

41.5

5×3

18

60

0.306

60

41.5

5×4

24

60

0.408

45

41.5

6×1

7.2

72

0.102

216

49.8

6×2

14.4

72

0.204

108

49.8

6×3

21.6

72

0.306

72

49.8

3 Design of LED Driver The LED driver is modeled as a closed-loop system as shown in Fig. 3. The sensor output corresponding to the output voltage signal is compared with reference input voltage Vref to obtain the error. The error signal is modulated to produce a corresponding control signal using a PI controller. The value of proportional constant (kp) is taken as 1 and that of integral constant (ki) as 500 for LED modules with 1.2 W chips. The transfer function of PI controller is given by Eq. 3.

Fig. 3 Block diagram of designed LED driver

Performance Study and Stability Analysis of an LED Driver

Gc = kp +

ki s

151

(3)

A gate pulse of required width is generated by a pulse-width modulation (PWM) generator, which regulates the output voltage and current of the buck-boost converter. The transfer function of the PWM generator (GPWM ) is given by Eq. 4, GPWM =

1 VM

(4)

where VM is the maximum value of the saw tooth carrier signal and is taken as 10 for LED modules modeled with 1.2 W chips. The transfer function of the buck-boost converter obtained with the averaging technique [7] is given by Eq. 5.   Vg DL s − (1 − D)2 R   Gvd (s) = (1 − D)2 LCs 2 + RL s + (1 − D)2

(5)

The loop gain of the proposed system is given by Eq. 6. T = 0.1GC .GPWM Gvd

(6)

The transfer function of the complete closed loop system is given by Eq. 7. GCL =

GC ∗ GPWM ∗ GVd d 1+T

(7)

4 Stability Analysis The stability analysis is performed using the system characteristics which are to be obtained from the bode plots of the loop gain of the proposed system. From the Bode plots using Phase margin test [8], stability of system can be analyzed. (1) Obtained Results The mathematical model of the LED driver operating the 6–24 W LED modules was tested in MATLAB for the steady-state stability at three input voltages (180 − 230 − 265 V). The phase margin obtained from the open loop bode plots of the LED driver are given in Table 3 and the variation of Phase Margin with Ns /Np ratio is shown in Fig. 4. It is observed that the phase margin (equal to 170 or 171 degrees) is greater than the required value of 76° (according to phase margin test) and remains almost constant for all Ns /Np ratio at all input voltages. The bode plot of the closed loop system for 21.6 W module (Ns /Np = 2) is shown in Fig. 5.

152

P. Ganguly et al.

Table 3 The phase margin of the LED driver at different input AC voltages Output power(w)

Ns /Np ratio

Phase margin (degree) at different input AC voltages (Vrms ) 265

230

180

9.6

2

171

171

171

6

5

171

170

170

12

2.5

171

171

171

18

1.66

171

171

171

24

1.25

171

171

171

7.2

6

170

170

170

14.4

3

171

171

170

21.6

2

171

171

170

14.4

1.33

171

171

171

19.2

1

171

171

171

Fig. 4 Variation of phase margin with Ns /Np ratio

5 Simulation Results The Simulink block diagram is shown in Fig. 6. The performance of designed LED driver is evaluated in MATLAB software and the results are given below in Table 4. The waveforms of input current, output current and output voltage for the designed LED driver obtained from simulation, for 9.6 W 4 × 2 LED module, are shown in Figs. 7, 8 and 9 respectively.

Performance Study and Stability Analysis of an LED Driver

Fig. 5 Bode plots of 21.6 W load for 6 × 3 series–parallel combination of LED module

Fig. 6 Simulation diagram

153

154

P. Ganguly et al.

Table 4 Electrical parameters of designed LED driver Load (W)

6

12

18

24

7.2

14.4

21.6

9.6

14.4

19.2

Series–parallel combinat ion of LED module

Input parameters Voltage (V)

Current (A)

Power (W)

5×1

265

0.1031

26.98

2.3

0.98

230

0.1067

24.33

1.8

0.99

5×2

5×3

5×4

6×1

6×2

6×3

4×2

4×3

4×4

THD (%)

Power factor

180

0.0724

12.87

6.2

0.98

265

0.1417

37.2

7.6

0.99

230

0.1468

33.45

9.3

0.99

180

0.1619

28.72

15.9

0.98

265

0.1662

42.55

25.1

0.96

230

0.1625

35.81

28.76

0.95

180

0.1974

33.7

33.17

0.93

265

0.191

47.45

36.3

0.94

230

0.2049

43.79

39.2

0.92

180

0.2358

38.72

44.74

0.94

265

0.1166

30.9

1.9

0.99

230

0.1224

27.96

1.5

0.99

180

0.0724

12.87

6.2

0.98

265

0.1395

35.54

26.51

0.96

230

0.1769

40.48

6.1

0.99

180

0.1953

34.81

13.21

0.99

265

0.1995

51.41

22.7

0.97

230

0.2133

47.34

26.51

0.96

180

0.2438

41.63

33.28

0.94

265

0.1117

29.17

8.8

0.98

230

0.1164

26.4

12.1

0.98

180

0.1288

22.68

19.29

0.97

265

0.1334

33.84

27.7

0.95

230

0.1371

29.6

35.31

0.93

180

0.1596

26.9

36.92

0.93

265

0.1576

38.36

41.51

0.99

230

0.1687

35.41

44.02

0.91

180

0.194

31.29

49.37

0.90

Performance Study and Stability Analysis of an LED Driver

155

Fig. 7 Input current waveform of LED driver for 9.6 W 4 × 2 LED module

Fig. 8 Output current waveform of LED driver for 9.6 W 4 × 2 LED module

5.1 Input Power Factor The variation of input power factor with the Ns /Np ratio is shown in Fig. 10. The input power factor of the LED driver is greater than 0.9 for all LED modules of 6–24 W and the input power factor increases with the increase in Ns /Np ratio.

156

P. Ganguly et al.

Fig. 9 Output voltage waveform of LED driver for 9.6 W 4 × 2 LED module

Fig. 10 Variation of the input power factor with Ns /Np ratio for 230 V

5.2 Total Harmonic Distortion (THD) THD of the designed LED driver is less than 50% for the entire output voltage range of 6–24 W. The variation of THD with the Ns /Np ratio is shown below in Fig. 11. From the figure it is shown that THD, expressed in percentage, decreases with the increase in Ns /Np ratio. It is less than 20% for Ns /Np ratio more than 2.

Performance Study and Stability Analysis of an LED Driver

157

Fig. 11 Variation of THD with Ns /Np ratio of the LED driver for 230V

5.3 Input Current Harmonics The values of odd harmonics are taken into consideration and compared with the recommended values of odd harmonics for class C type equipment (Lighting equipment) in Standard EC 61000-3-2 (Harmonics Standards Overview) [9]. It is observed that for third harmonic the obtained values are less than the recommended values, so it complies with the standard for all values of Ns /Np ratio. But for 5th, 7th, 9th, 11th harmonics, it is observed that when Ns /Np ratio is greater than or equal to 2, the obtained values complied with the prescribed standard values. The variation of input current third harmonic with Ns /Np ratio is shown in Fig. 12.

Fig. 12 Variation of the 3rd harmonic of input current with Ns /Np ratio

158

P. Ganguly et al.

6 Conclusions In this work, an LED driver has been designed and simulated in MATLAB Simulink which can operate 6–24 W LED modules designed with 1.2 W LED chip. The steadystate stability of the designed LED driver has been obtained from its mathematical model using phase margin test and the LED driver is found to be stable for all LED loads in the voltage range of 180–265 V. The electrical performance of the designed LED driver has been obtained through simulation. From the simulation results, it is observed that the input power factor is greater than 0.9 for all LED modules, but the THD is greater than 20% for LED modules having Ns /Np ratio less than 2. The 3rd harmonic complies with relevant International standards for all Ns /Np ratios, but the other odd harmonics comply only for Ns /Np ratio greater than or equal to 2.

References 1. Zhang J, Jiang T, Xu L, Wu X (2013) Primary side constant power control scheme for LED drivers compatible with TRIAC dimmers. J Power Electron 13(4):609–618 2. Jha A, Singh B (2017) Bridgeless ZETA PFC converter for low voltage high current LED driver. In: 6th international conference on computer applications in electrical engineering recent advances (CERA) 3. Sulistiyanto N, Rifan M, Setyawati O (2015) Design and prototype of LED driver based on the buck converter using FPGA module. In: International conference on quality in research 4. Saxna AR, Kulshrestha A (2017) Universal bus front end PFC fourth-order buck converter as LED drivers. In: TENCON IEEE Region 10 conference 5. Gacio D, Alonso M, Calleja A, Garcia J, Rico-Secades M (2011) A universal-input single-stage high- power—factor power supply for HB-LEDS based on integrated Buck-Flyback converter. IEEE Trans Ind Electron 58(2):589–599 6. Gupta V, Basak B, Ghosh K, Roy B (2017) Stability analysis of a universal LED driver. In: IEEE calcutta conference (CALCON) 7. Zhou X, He Q (2015) Modeling and simulation of Buck-Boost converter with voltage feedback control. In: MATEC web of conferences, vol 31, p 10006 8. Erikson RW, Fundamental of power electronics. 2nd edn. Springer 9. IEC 61000-3-2 Harmonics Standards Overview

Instrumentation for Wireless Condition Monitoring of Induction Machine Soumyak Chandra, S. Saruk Mohammad and Rajarshi Gupta

Abstract In this paper, we describe the instrumentation setup for wireless condition monitoring of induction machine in a laboratory environment. Four parameters, viz., stator voltage, current, body temperature, and rotor speed were sensed and fed to a dedicated hardware unit (DHU). This DHU was operated using commands from a desktop computer, acting as an operator station. An Arduino Uno board was used as the data collection unit of the DHU and interfacing to the ZigBee module. Data packets with all four sensed parameters of 5 s duration was continuously collected and transferred to the operator station for real-line display in a graphical user interface. All parameters had less than 1% error in their measurements. Keywords Condition monitoring · Induction machine · Signal conditioning · ZigBee · Graphical user interface

1 Introduction Condition monitoring (CM) is a process to check up the health of a system by monitoring some parameters in order to identify their significant change by which any developing fault can be detected. CM of electric machinery can significantly reduce the cost of maintenance and the risk of unexpected failures by allowing the early detection of potentially catastrophic faults. Online CM uses measurements taken while a machine is operating, to determine if a fault exists. Different types of

S. Chandra (B) · S. Saruk Mohammad · R. Gupta (B) Electrical Engineering Section, Department of Applied Physics, University of Calcutta, Kolkata, India e-mail: [email protected] R. Gupta e-mail: [email protected] S. Saruk Mohammad e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_13

159

160

S. Chandra et al.

sensors are used to measure output signals, followed by various signal processing techniques to extract particular features which are sensitive to the presence of faults. Finally, in the fault detection stage, a decision needs to be made as to whether a fault exists or not. A vast research literature is available for CM of various electrical, mechanical, and electronic systems, spanning from power transformers [1, 2], rotating machines like induction machines [3, 4], alternators in power stations [5], human health [6], massive structural elements like bridges [7, 8], wind turbines [9], high-rise buildings, air pollution [10], river health condition [11], and crop condition [12] for agricultural applications. The four components in CM are: sensors and signal conditioning, a communication framework, data acquisition in the host device and data analysis for decision-making. The sensor array and signal conditioning forms the part of signal collection from the system being monitored. This consists of various signal refinement circuits that collects the desired signal(s) only. For remote data collection, as in case of wireless monitoring, a communication framework is essential that connects the field sensors to the host through a wireless channel. For distributed monitoring, wireless sensor networks play a vital role in data aggregation and presentation to the host [13, 14]. Finally, a software in the host computer is used for data collection and signal analysis of the received data. For induction machines, the main causes of failure are: short air gap leading to air gap stability problems, slip ring and brush gear problems, and high starting current leading to thermal and fatigue failure. As far as worldwide research in induction machine failure is concerned, most of the publications addressed the following issues, viz., rotor cage broken bars, discharge activity under medium and high-voltage stator windings, stator winding faults (excluding discharge activity), and bearing faults. In [15] the authors use dedicated embedded systems for signal acquisition of different electrical parameters like voltage, current, power factor, frequency, etc., from multiple induction machines from a centralized computer using serial communication (RS-232) and (RS-485). Another aspect in CM is use of data analysis tools for abnormality detection. To avoid errors in human judgment and reduce operator dependability, different artificial intelligence based methods have been utilized for automatic diagnosis of machine health [16, 17]. In such analysis, a number of features are extracted from the voltage, current, and temperature data which describe a particular profile characteristics of the machine state. A classifier then provides binary decision to infer the state as “normal” or “abnormal”. In the current work, we describe the sensing and data collection framework for wireless condition monitoring of induction machine(s), excluding the computerized data analysis. The rest of the paper is organized as follows. Section 2 provides detailed description of the circuits and systems used for the laboratory based prototype used. Section 3 shows the experimental results within a laboratory environment. Section 4 states the main outcome of the work and possible future scope of work.

Instrumentation for Wireless Condition …

161

2 Methodology Block diagram of the system layout is shown in Fig. 1. The whole system is divided into two parts, viz., machine-end dedicated hardware unit (DHU) and computer as the local host or, operator station. The DHU contains the sensor interfacing units, an Arduino Uno board, and a ZigBee module. In this work, we measure the stator current and voltage, shaft rotational speed, motor body temperature. All of these sensing units have their respective signal conditioning circuits to present the output in the range 0–5 V to the Arduino Uno input. Figure 2 shows the DHU layout. The voltage and current signals are converted to DC before being fed to analog to digital converter (ADC) of the microcontroller. The sampling rate for voltage, current, temperature and shaft speed are kept at 20 Hz, 20 Hz, 10 Hz, and 10 Hz, respectively. The monitored parameters are fed to the ATmega328 microcontroller of the Arduino Uno board. In the receiving end, the computer acquires the data through a ZigBee module attached to the USB port. A graphical user interface (GUI) was developed in the computer to monitor motor parameters from a remote location. The signal conditioning circuits for rotor speed, temperature, stator current, and voltage sensing are ∼



ZigBee Module

ZigBee Module

… Monitoring Station

Dedicated Hardware Unit

Dedicated Hardware Unit

Fig. 1 Block diagram of the proposed system layout

Power Supply Potential Transformer

Signal Conditioning Unit

Current Transformer

Signal Conditioning Unit

Proximity Sensor

Signal Conditioning Unit

Fig. 2 Block diagram of the DHU

Arduino UNO ATmega328

LM35 Temp. Sensor

Signal Conditioning Unit

162

S. Chandra et al.

shown in Fig. 3a–d respectively. The speed measurement of the motor was done by a variable reluctance type proximity sensor along with a soft-iron nut attached to the rotating shaft. It generates a pulse train the frequency of which varies with the speed. LM2907 is a frequency to voltage converter using the principle of a charge pump. In this work, temperature sensing was done by a LM35 IC, which provides a sensitivity of 10 mV/°C. It is powered by a +5 V supply. The raw sensor output is fed to a differential amplifier. Since the measured temperature range is 25–120 °C, the first stage of the signal conditioning is zero adjustment by a 10 K trim pot. The gain adjustment was done by the feedback resistance 4.7 K. The stator current and voltage was measured by a current transformer (CT) and potential transformer (PT) respectively, followed by their signal conditioning circuits, both utilizing a precision full-wave rectifier.

Fig. 3 Signal conditioning circuit for a speed; b temperature; c stator current; and d stator voltage measurement

Instrumentation for Wireless Condition … Machine ID 1 byte

163

Current

Voltage

Temperature

Speed

CRC

100 bytes

100 bytes

50 bytes

50 bytes

1 byte

Fig. 4 Packet structure of the transmitted packet

At startup, the operator station provides facility to know the presence of the machines through a global broadcast message using the GUI of the operator station. The DHUs respond back with a short message containing their IDs, which are refreshed immediately on the GUI. The operator can then select only one machine at a time and send a command with data request. Hence, the communication between the DHU and Operator station is essentially a peer-to-peer, or conforms to star topology. When a particular DHU receives the “data send” command, it collects the data from all sensors of the particular machine and transmits data packets at every 5 s frame through ZigBee unit. The communication between the ATmega328 controller and ZigBee is purely 8-bit serial universal asynchronous receiver transmitter (UART). Detailed packet structure is given in Fig. 4. The received data is unpacked and plotted in real time on the respective panel of the GUI.

3 Results and Discussions The proposed system was hardware implemented in laboratory scale, keeping a distance of 10 feet between the induction machine unit and operator station. However, the distance could be enhanced up to 30 feet without loss of data. Figure 5 shows the physical layout DHU showing all sensors, and signal conditioning units. Figure 6 shows a screenshot of the operator station GUI with acquired data plot. Initially, the “Open” push button switch is used to open the serial port and for scanning the presence of all DHUs. Shortly, the machine ID panel is flushed with the available DHUs as green color and rest (unavailable) with red colour. The operator can choose the machine ID from the radio button of the available DHUs, and click on “Acquire” push-button. The DHU starts sending the data packets which are segregated and plot in real-time on the four graphic panel, which gets refreshed after every 5 s. The operator can save the data collected using the “Save Data” push button. If the operator wants to refresh the GUI with blank panels, he/she can use “RESET” push button. The ongoing acquisition session can be terminated using the “Close” push-button, which closes the serial port. Without “Reset’, clicking the ‘Close’ push button will only close the serial port, not stopping the incoming acquired data stream from the ZigBee module. Since the acquired data to be shown in the direct units of voltage, current, RPM and °C respectively in the operator station, calibration characteristics of the individual sensors were determined offline. Here, all signal condition circuits linearly translate the input to a voltage value, the characteristic equation can be represented as

164

S. Chandra et al.

Variac

Induction machine

DHU

Temperature Sensor Arduino Uno board ZigBee module Fig. 5 Physical layout of the DHU

Fig. 6 Layout of the operator station

y = mx + c

(1)

where y is output in volts, x is the parameter measured by a standard multimeter (Fluke 17B plus), a digital tachometer (accuracy of 0.1 rpm), and a mercury in glass thermometer. Now, this y is translated in a scale of 0–102310 in the microcontroller. The microcontroller transmits these raw data in the form of data packets with an error checking code.

Instrumentation for Wireless Condition … Fig. 7 Calibration characteristics of the different sensing units under different condition

165

166

S. Chandra et al.

Upon reception, the value of x in respective unit was directly converted into the respective parameters (volt, amps, RPM and °C) in the operator station software before displaying into the real-time graphical display. Figure 7 shows the calibration characteristics of the all parameters monitored in this study. Figure 7a, b shows the speed versus load characteristics under two stator voltages, viz., 220 and 160 V the measurements shows an average percentage error of 0.34% and 0.31% respectively under 220 V and 160 V respectively. Figure 7c shows the calibration characteristics of the CT, where the x-axis shows the measured voltage across CT burden and y axis shows DC output of signal conditioning. Figure 7d, e shows the calibration characteristics of the PT for 3 kg load and no load condition respectively, with an error figure of 0.81% and 0.99%, respectively.

4 Conclusions This paper describes the instrumentation setup for wireless condition monitoring scheme using ZigBee wireless communication. One advantage of the scheme is that an operator can readily access the individual machines from the operator station using the GUI. The experimental results show fairly accurate results, with less than 1% error in speed, temperature, and voltage. For Laboratory environment star network topology between the DHUs and the operator station is sufficient which is also most simple, at the cost of higher energy consumption. For outdoor measurements, use of mesh network could consume overall less energy. However, the data routing would be more sophisticated. Acknowledgements We acknowledge Sri Sourav Biswas, Technical Asst., Workshop, Department of Applied Physics, University of Calcutta for technical assistance in the sensor fixation and board placement in the hardware setup. We also acknowledge Dr. Kaushik Das Sharma, Associate Professor, Electrical Engineering Section, and Dr. Samarjit Sengupta, retired Professor of the same department for providing valuable suggestions and technical guidance to accomplish the work.

References 1. Kogan VI, Fleeman JA, Provanzana JH, Shih CH (1988) Failure analysis of EHV transformers. IEEE Trans Power Deliv 3(2):672–683 2. Leibfried T (1998) Online monitors keep transformers in service. IEEE Comput Appl Power 11(3):36–42 3. Thomson WT, Rankin D, Dorrell DG (1999) On-line current monitoring to diagnose airgap eccentricity in large three-phase induction motors—industrial case histories verify the predictions. IEEE Trans Energy Convers 14:1372–1378 4. Stone G, Kapler J (1998) Stator winding monitoring. IEEE Ind Appl Mag 4(5):15–20 5. Albright DR, Albright DJ, Albright JD (1999) Flux probes provide on-line detection of generator shorted turns. Power Eng 103(9):28–32

Instrumentation for Wireless Condition …

167

6. Roy S, Gupta R (2014) Short range centralized cardiac health monitoring system based on ZigBee communication. In: 2014 IEEE global humanitarian technology conference—South Asia satellite, GHTC-SAS 2014, pp 177–182 7. Wenzel H (2009) Health monitoring of bridges. Wiley 8. Ko JM, Ni YQ (2005) Technology developments in structural health monitoring of large-scale bridges. Eng Struct 27(12):1715–1725 9. Yang W, Court R, Jiang J (2013) Wind turbine condition monitoring by the approach of SCADA data analysis. Renew Energy 53:365–376 10. Tsujita W, Yoshino A, Ishida H, Moriizumi T (2005) Gas sensor network for air-pollution monitoring. Sens Actuators B Chem 110(2):304–311 11. Norris RH, Hawkins CP (2000) Monitoring river health. Hydrobiologia 435:5–17 12. Doraiswamy PC, Hatfield JL, Jackson TJ, Akhmedov B, Prueger J, Stern A (2004) Crop condition and yield simulations using landsat and MODIS. Remote Sens Environ 92(4):548–559 13. Othman MF, Shazali K (2012) Wireless sensor network applications: a study in environment monitoring system. Proc Eng 41:1204–1210 14. Pandian PS et al (2008) Wireless sensor network for wearable physiological monitoring. J Netw 3(5):21–29 15. Datta J, Chowdhuri S, Bera J, Sarkar G (2012) Remote monitoring of different electrical parameters of multi-machine system using PC. Meas J Int Meas Confed 45(1):118–125 16. Wang Z, Liu Y (1998) ANN and expert system tool for transformer fault diagnosis. IEEE Trans Power Deliv 13(4) (vol 2):1261–1269 17. Wang Z, Liu Y, Griffin PJ (2000) A combined ANN and expert system tool for transformer fault diagnosis. In: 2000 IEEE power engineering society, conference proceedings, pp 1261–1269

Solar PV Battery Charger Using MPPT-Based Controller Shreya Das, Avishek Munsi, Piyali Pal, Dipak Kumar Mandal and Sumana Chowdhuri

Abstract Present circumstances lead a major challenge to the non-renewable sources of energy, causing more need of renewable energy. The common characteristics of renewable energy except for hydroelectricity are low energy density and high resource dispersion. Moreover, wind, solar, and ocean energies are stochastic and intermittent resources but solar energy is available in an ample amount almost in every strata of the world and is an effective renewable source of energy. Here we develop a prototype of large-scale system for feeding local domestic loads with solar energy and to store the energy for using, during peak hours when our traditional energy sources are not able to satisfy the demand of energy needed by the domestic loads. In our small scale approach at laboratory environment, instead of using solar photo-voltaic module, we used solar photovoltaic simulator as source to generate electricity. The source voltage is then converted by using a DC–DC buck type converter to a required value for charging a 12 V lead-acid battery which is used as an energy storage element. Here we implemented the Maximum Power Point Tracking algorithm of perturb and observe technique to extract maximum power and to transfer the extracted energy to our desired storage element by developing a program which is interfaced with hardware using Arduino Uno. Keywords MPPT technique · Perturb and observe technique · DC–DC buck converter · Energy storage element S. Das (B) · A. Munsi (B) · P. Pal · D. K. Mandal · S. Chowdhuri Electrical Engineering, Department of Applied Physics, University of Calcutta, Kolkata, India e-mail: [email protected] A. Munsi e-mail: [email protected] P. Pal e-mail: [email protected] D. K. Mandal e-mail: [email protected] S. Chowdhuri e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_14

169

170

S. Das et al.

1 Introduction India is a tropical country receiving direct rays of sunlight almost throughout the year. The solar photovoltaic (PV) can convert that sunlight into electrical energy. So solar energy can fulfill the growing electrical energy demand and energy security. Moreover, as solar energy is clean and free from emissions so it is environmentfriendly. But the solar energy is variable in nature and available only during daytime. So, an efficient controller is required to extract the maximum available power for utilization. The extracted power can be directly fed into the distribution grid. But it is only possible during daytime. The energy must be stored for use at night when sun is absent. Here a 12 V lead-acid battery has been used for storage purpose. So instead of feeding the energy directly to the grid this energy can be stored. It can overcome the duck curve problem of the current PV system. The solar energy harnessed during the day can be used at night when the consumption is highest. But the output power of PV panels changes with changing irradiance and temperature. The output characteristics of PV system vary nonlinearly and are influenced by solar irradiance and ambient temperature. The daily solar irradiation has abrupt variations throughout the day due to sudden cloud coverage or shadow of any object like birds, trees, banners, etc. Due to which the maximum power point (MPP) of PV array changes continuously. Then the system’s operating point should also be changed considerably so as to maximize the energy produced. Ramaprabha et al. [1] explains before how maximum power point tracking is an efficient way to track maximum power using GA-optimized Artificial Neural Network algorithms and boosted the source voltage using Boost converter to operate desirable loads but there was no storage problem solution. Hadji et al. [2] developed genetic algorithms to carry out maximum power point tracking (MPPT) based on the cell model and compares it with conventional Perturb and Observe and Incremental conductance algorithms to overcome the problems arise there due to rapidly changing of atmospheric conditions. Also they resolved the problem of oscillations around MPP [3]. Our objective is to feed the energy to load side during peak hours when our traditional energy sources are not able to satisfy the demand of energy needed by the domestic loads. Also by charging a lead-acid battery, we introduce a way to store energy for future use, as discussed above.

2 System Development Methodology 2.1 Solar Charger Schematic Figure 1 shows that the output from the solar panel (maximum of 18 V) is stepped down to 14 V using a DC–DC buck-type converter. Here, the duty cycle of the PWM

Solar PV Battery Charger Using MPPT-Based Controller

171

Fig. 1 Proposed system block diagram

signal fed to the switch of the converter is controlled by using a MPPT controller [4] which is done with the help of an Arduino Uno. The output voltage and current from the solar panel is sensed by a sensor circuit and is fed to the Arduino Uno. PWM signal is produced as per the MPPT algorithm programmed in the Arduino, which is then used to switch the buck converter. Finally a 12 V lead-acid battery is used as load. The battery is charged [5] with the output power of the DC-DC buck type converter.

2.2 Maximum Power Point Tracking (MPPT) There are different types of MPPT techniques, among which we worked on the Perturb and Observe (P&O) technique [6]. The Perturb and observe algorithm [7] operates on the array voltage and current. It compares the present output power with the previous output power after periodic perturbations. If the array operating voltage changes such that the output power increases, the algorithm keeps on changing the operating voltage in the same manner. Whereas, if the output power decreases, the algorithm changes the operating point in the opposite direction. In the next perturbation cycle the algorithm repeats itself in the same way. Figure 2 shows the PV and VI general characteristics of solar module. Pmax indicates the maximum power point. The algorithm we used to develop our program is conventional Perturb and Observe method and it is shown in Fig. 3. Arduino Uno has been used for programming of the controller in P&O technique. Digital implementations make the technique less expensive than analogical implementations. Extra circuitry for safety purpose is eliminated effectively by the use of Arduino, thus simplifying the overall circuit. It is easier and much reliable to measure voltage and current. Expensive and rare irradiance and temperature sensors are not required. A voltage divider circuit is used to sense voltage and a 1  resistance is used to sense current which further makes the circuit cheaper and less bulky than using expensive and bulky current and voltage sensors. Unlike SC (Short Current

172

Fig. 2 Characteristics of solar PV module

Fig. 3 Perturb and observe algorithm

S. Das et al.

Solar PV Battery Charger Using MPPT-Based Controller

173

Fig. 4 Change in current versus voltage characteristics of PV module with sudden decrease in irradiance

Pulse Method) and OV (Open Voltage Method) techniques, P&O technique does not require additional static switches. But for partial shading condition the output power oscillates around the MPP as shown in Fig. 4 (where the operating point oscillates between A and B due to sudden change in irradiance) resulting in loss of power in the PV system. P&O technique fails [8] during rapidly changing atmospheric conditions.

2.3 DC–DC Converter Normally for the Photovoltaic system with battery as load, the MPP of commercial PV module is set beyond the charging voltage of battery for most combinations of irradiance and temperature. In such case, buck converter is used to decrease the voltage of the PV module to the charging voltage of the batteries. But if the MPP [9] goes below the battery charging voltage due to low irradiance and high temperature then the buck converter is of no use. Thereafter we need an additional boost capability. In our work we are using a buck type DC–DC converter as shown in Fig. 5. A buck converter [10] is designed by using MOSFET as switching device for high-frequency switching action. An inductor, a capacitor, a load and a freewheeling diode is also used. The gate pulse is supplied to the MOSFET from the Arduino where the MPPT algorithm is programmed. As the MOSFET is switched on by providing gate pulse, the diode remains open as it is in the reverse biased condition. Current flow increases through the inductor and the inductor stores energy. When the MOSFET is switched off, the diode is forward biased. Then current flowing through the inductor decreases Fig. 5 DC–DC buck converter schematics

174

S. Das et al.

and the stored energy is thereafter released through the diode. L=

V0 (Vs − V0 ) I L f s Vs

(1)

I L 8V0 f s

(2)

C= where

V0 = Output voltage of the converter = 14 V Vs = Input voltage of the converter = 18 V (PV module’s output voltage) IL = Ripple current = 0.19 A fs = Switching frequency = 4 kHz Vo = Ripple voltage at the output = 1.249 × 10−3 .

3 Simulation and Hardware Implementation of the Design 3.1 MATLAB Simulation The proposed system as Fig. 1 has been simulated in MATLAB simulink. As discussed earlier the voltage has been stepped down from 18 V to 14 V by using a step down chopper (Buck Regulator) shown in Fig. 6 to charge a lead-acid battery. The switching frequency of the buck converter is taken as 4 kHz with duty cycle of 80%. The parameter of buck converter has been selected after observing the suitable case for which the ripple is very less and neglected.

Fig. 6 Hardware implementation of the DC–DC buck converter

Solar PV Battery Charger Using MPPT-Based Controller

175

3.2 Hardware Implementation The complete system circuitry is shown in Fig. 7. The input from the solar module is first sensed by voltage and current sensors. The sensed value is fed to the Arduino for the MPP tracking. The Arduino, following the P&O algorithm stored in it, sends pulse to the gate of the converter switch, which is a MOSFET, via an opto-coupler. The buck converter then stepped down the voltage from the solar module to charge a 12 V Lead-Acid battery (Fig. 8).

Fig. 7 Proposed circuit diagram

Fig. 8 Hardware implementation

176

S. Das et al.

3.3 Voltage and Current Sensor The sensor circuit, as shown in Fig. 9, senses the voltage and the current from the solar panel and feed them to the Arduino analog inputs for implementation of the MPPT algorithm. As the Arduino has maximum input voltage level of 5 V so the sensed values are reduced to 5 V. (1) Voltage sensor: A potential divider is used as voltage sensor. It drops the solar panel voltage from maximum 21 V (open circuit voltage of PV module) to 5 V. (2) Current sensor: As we used parallel connected 3 solar panels so the maximum current will be thrice the rated current of each solar panel (3 × 0.6 A = 1.8 A). A 1 Ω shunt is used such that the voltage across the shunt is 1.8 V (max). This voltage is amplified using OP-Amp LM324 (operating it in non-inverting mode) to almost 5 V and fed to the analog input of the Arduino.

3.4 Arduino Program Pseudocode Step 1: Step 2: Step 3: Step 4: Step 5: Step 6: Step 7: Step 8:

Set Baud rate at 9600 and converter switching frequency at 4 kHz. Set the initial duty of gate signal to the MOSFET converter 60%. At a time interval of 250 µs the sensed values of current and voltage are stored in variables. With these current and voltage, power is calculated. If the power is increasing go to step 6 else go to step 7. If the voltage is increasing go to step 8 else go to step 9. If the voltage is increasing go to step 9 else go to step 8. Increase duty of the square wave signal by 10% fed to the gate of the MOSFET; go to step 10.

Fig. 9 Sensor circuit

Solar PV Battery Charger Using MPPT-Based Controller

177

Step 9:

Decrease duty of the square wave signal by 10% fed to the gate of the MOSFET. Step 10: Update history by storing the present values of current, voltage and power to the variables storing the respective previous values; go to step 3.

4 Results and Discussion The output of the buck type DC–DC converter in MATLAB simulation is shown in Figs. 10 and 11. Figure 10 shows current versus time plot and Fig. 11 shows the output voltage versus time plot which shows that both the current and voltage at the output has minimum ripple. This voltage is supplied to the load (12 V lead-acid battery). Figure 12 shows the gate pulse which is fed to the switch of the MOSFET.

Fig. 10 Current versus time plot of output of buck converter

Fig. 11 Voltage versus time plot of output of buck converter

178

S. Das et al.

Fig. 12 Voltage versus time plot of gate signal to MOSFET of buck converter

The MATLAB simulation of a PV panel of rated voltage 17.71 V. The output current supplied from the PV panel is 0.6248 A. So 3 solar panels of such rating has to be connected in parallel to receive the required supply of 17.71 V and (3 × 0.6 A = 1.8 A). The power (P) versus voltage (V) and voltage (V) versus current (I) characteristics of the solar module for different illumination is shown in Figs. 13 and 14, respectively. Data to plot these curves are taken from a PV panel of considered rating (hardware). In the figures given below insolation I is greater than insolation II, which is again greater than insolation III. The maximum power point differs for different illumination of light (Figs. 13 and 14).

Fig. 13 Power versus voltage characteristics

Solar PV Battery Charger Using MPPT-Based Controller

179

Fig. 14 Current versus voltage characteristics

The final result is shown by a computer screen shot in Figs. 15 and 16 in the next page, which is done using PV simulator. The result shows maximum efficiency of maximum power point tracking using Perturb and Observe technique. Figures 17, 18 and 19 shows the change in duty cycle of the gate signal supplied to the MOSFET of the buck converter for changing voltage of the solar module (hardware). The output voltage of the buck converter changes linearly with the input voltage from the solar module and saturates to 14 V, which is supplied to the battery. We performed our P&O algorithm in Arduino and used PV simulator to track the maximum power point.

Fig. 15 Result showing maximum efficiency of MPPT in a PV simulator

180

Fig. 16 MATLAB simulation of PV module Fig. 17 71.5% duty cycle of gate signal for Vin = 19.58 V

Fig. 18 80.8% duty cycle of gate signal for Vin = 17.35 V

Fig. 19 84.6% duty cycle of duty cycle for Vin = 16.5 V

S. Das et al.

Solar PV Battery Charger Using MPPT-Based Controller

181

5 Conclusion An efficient photovoltaic battery charging system with the capability of tracking the maximum power point using perturb and observe technique at a very low cost is designed and implemented. By the use of an Arduino Uno, the complexity of writing hexadecimal code in microcontrollers is avoided, which has simplified the software part of the project. The battery overcharging protection is also programmed in the Arduino Uno. This reduced the expense of extra circuitry need for protection of the storing element. Instead of using expensive sensors for current and voltage sensing, cheap and simple circuits using resistances and an op-amp are used, thereafter making the circuitry less bulky. The buck converter is designed according to the required ratings using capacitors, MOSFET, a freewheeling diode and a handmade crude inductor. P&O technique can track the MPP avoiding the need of static switches, as is required in SC and OV techniques. Thus the MPP is reached using a very simple circuitry at a very less expense. The perturb and observe technique is quite an effective way of tracking the maximum power point. But to overcome the drawbacks of frequent change in solar irradiance as already explained earlier can be overcome by using the incremental conductance technique. Conclusion can be made from the result obtained by the PV simulation (as shown in Fig. 15) the MPP is detected at a good efficiency using the P&O technique for a slow change in illumination. Acknowledgements This work has been performed at Department of Applied Physics in University of Calcutta with the support of DST, Government of India Funded Project.

References 1. Ramaprabha R, Gothandaraman V, Kanimozhi K, Divya R, Mathur BL (2011) Maximum power point tracking using GA-optimised artificial neural network for solar PV system. Newport Beach, CA, USA, pp 264–268, 7 Mar 2011 2. Hadji S, Krim F, Gaubert J (2011) Development of an algorithm of maximum power point tracking for photovoltaic system using genetic algorithm. Tipaza, Algeria, pp 43–46, 27 June 2011 3. Esram T, Chapman P (2007) Comparison of photovoltaic array maximum power point tracking techniques. IEEE Trans Energy Conversat 22:439–449 4. Faranda R, Leva S (2008) Energy comparison of MPPT techniques for PV systems. WSES Trans Power Syst 3:446–455 5. Horkos P, Yammine E, Karami N (2015) Review on different charging techniques of lead-acid batteries. In: Proceedings of the third international conference on technological advances in electrical, electronics and computer engineering (TAEECE), pp 27–32 6. Altas I, Sharaf A (1996) A novel on-line MPP search algorithm for PV arrays. IEEE Trans Energy Conversat 11:748–754 7. Liu X, Lopes L (2004) An improved perturbation and observation maximum power point tracking algorithm for PV arrays. In: Proceedings of the power electronics specialists conference, vol 3, pp 2005–2010

182

S. Das et al.

8. Femia N, Petrone G, Spagnuolo G, Vitelli M (2005) Optimization of perturb and observe maximum power point tracking method. IEEE Trans Power Electron, pp 963–973 9. Femia N, Lisi G, Petrone G, Spagnuolo G, Vitelli M (2008) Distributed maximum power point tracking of photovoltaic arrays. Novel approach and system analysis. IEEE Trans Ind Electron 55:2610–2621 10. Ortiz-Rivera EI (2008) Maximum power point tracking using the optimal duty ratio for DC-DC converters and load matching in photovoltaic applications. IEEE, pp 987–991

Comparative Study on Simulation of Daylighting Under CIE Standard Skies for Different Seasons Abhijit Gupta and Sutapa Mukherjee

Abstract This paper deals with the comparative study on daylight availability on horizontal working plane of a simulated room under identified CIE (International Commission on Illumination) Standard skies prevailing in Roorkee, India for the three seasonal conditions, viz. equinox, summer, and winter solstices. Daylight coefficient method (DC) and finite element method (FEM) have been applied to develop computer programs in MATLAB environment for this simulation. Here daylight availability is predicted for a room with single-sided window of opening areas 20% of floor area and sill height 1 m from floor with eight cardinal window orientations. Analysis revealed that during summer the amount of daylight availability is maximum as the sun shines directly on the Northern Hemisphere during summer. Polar axis of earth is tilted 23.5° to the orbital plane. Differential and changing illumination pattern on earth for different seasons is due to combinations of rotation, revolution, and tilt of polar axis. Keywords CIE standard skies · Sky component computation · Daylight coefficient method · Finite element method · Rotation and revolution of earth

1 Introduction Daylight being natural light is available free of cost as well as necessary for maintaining circadian rhythm of humans. The solar heat gain associated with daylight penetration may be advantageous or disadvantageous to maintain thermal comfort to the occupants depending on local climatic condition. The available amount of daylight on horizontal working plane in indoor area can be predicted by computing the value of illuminance on array of grid points on the same. Here the source of daylight is basically the sky and its luminance pattern is dynamic in nature. However,

A. Gupta · S. Mukherjee (B) Department of Electrical Engineering, B. P. Poddar Institute of Management & Technology, Kolkata, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_15

183

184

A. Gupta and S. Mukherjee

luminance pattern of external building facade and ground also act as potential source of daylight in presence of large obstruction. SSLD (Standard Sky Luminance Distribution) model is required for calculating availability of daylight during different seasons. This model was published by CIE (International Commission on Illumination) in 2003 [1]. Relevant sky type for any location must be identified before using this SSLD model. For this, one of the important methods is Daylight Coefficient (DC) method [2]. In India at Roorkee, the relevant sky type for different sky types for different seasons are identified due to availability of monthly average hourly sky luminance data under IDMP (International Daylight Measurement Program) [3–7]. Using DC method, the available daylight simulation on the horizontal working plane of a room is being calculated with the identified sky [1, 2, 8–10]. Dynamic pattern of daylight distribution on the working plane is simulated by the computation program developed in MATLAB environment. Finite Element Method (FEM), based on theory of Radiosity, is applied to compute interreflection, i.e., indirect component of illuminance distribution. Room is considered as empty and window is unobstructed [4, 11]. The outcome of this simulation can be utilized during daylighting design stage for room of any dimension and surface reflectance to predict daylight availability at different seasons. Keeping the identified sky type fixed, the variation of sky luminance pattern is considered taking different local time and corresponding zenith luminance (Lz ). The effectiveness of solar radiation to produce daylight is measured by the ratio between illuminance (Lux) and irradiance (W/m2 ) and this ratio is known as daylight efficacy. There are established daylight efficacy models formulated for different climatic conditions; out of them Perez daylight efficacy model is mostly accepted one. This daylight efficacy model is the mathematical one and is used to get either global or diffuse illuminance value from the corresponding global and diffuse irradiance values. The applicability of Perez daylight efficacy model is experimentally validated for the climatic condition of Kolkata [12, 13]. The use of Perez daylight efficacy model is beyond the scope of present work.

2 CIE Sky Model: SSLD In 2003, CIE published Standard Sky Luminance Distribution (SSLD) model [7]. It provides relative spatial luminance distribution for fifteen CIE Standard General Skies. The luminance of any sky element specified by γ and α is modeled based on the theory of sunlight scattering within the atmosphere and expressed by the product of two different exponential functions viz, gradation function ϕ and indicatrix function given by [7]   (i)  π2 − γ and

Comparative Study on Simulation of Daylighting Under …

185

(ii) f (χ) Thus the expression of luminance of sky element is given by Eq. (1) L γ α = f (χ) ∗ 

π 2

−γ

 (1)

The zenith luminance Lz can be obtained from Eq. (1) by putting the value of γ = π/2. Then   χ = π2 − γs and thus Lz = f

π 2

 − γs ∗ Φ(0)

(2)

The CIE SSLD is the sky luminance distribution relative to the zenith luminance and is represented by Eq. (3).   f (χ) ∗  π2 − γ Lγ α  = π Lz f 2 − γs ∗ Φ(0)

(3)

L γ α = luminance (cd/m2 ) of any sky element L z = zenith Luminance (cd/m2 ) χ = scattering angle between the sun and sky element γs and αs = altitude and azimuthal angles of the sun Here angles are in radian and shown in Fig. 1. Zenith sky luminance (Lz ) values are computed for different local times and three different seasons from the following formula. These values are tabulated in Table 1. L z = Asinγs + 0.7(Tv + 1)

Fig. 1 Angles defining the position of the sun and a sky element [14]

(sinγs )C + 0.04Tv (cosγs ) D

(4)

186

A. Gupta and S. Mukherjee

Table 1 Lz values for three different seasons for five different local time Time (h)

8 am

Season

Lz (kcd/m2 )

10 am

12 noon

2 pm

4 pm

Summer Equinox

3.4

11.8

30.6

11.8

4.2

2.1

4.8

7.7

5.5

Winter

2.4

1.1

3.1

4.3

4.3

1.4

Table 2 Parameters applied for descriptor calculations in absolute units [14] Sky type

Sky code

Parameters A1

A2

C

D

E

1

I.1

54.63

1.00

0.00

0.00

2

I.2

12.35

3.68

0.59

50.47

3

II.1

48.30

1.00

0.00

0.00

4

II.2

12.23

3.57

0.57

44.27

5

III.1

42.59

1.00

0.00

0.00

6

III.2

11.84

3.53

0.55

38.78

7

III.3

0.957

1.790

21.72

4.52

0.63

34.56

8

III.4

0.830

2.030

29.35

4.94

0.70

30.41

9

IV.2

0.600

1.500

10.34

3.45

0.50

27.47

10

IV.3

0.567

2.610

18.41

4.27

0.63

24.04

11

IV.4

1.440

−0.750

24.41

4.60

0.72

20.76

12

V.4

1.036

0.710

23.00

4.43

0.74

18.52

13

V.5

1.244

−0.840

27.45

4.61

0.76

16.59

14

VI.5

0.881

0.453

25.54

4.40

0.79

14.56

15

VI.6

0.418

1.950

28.08

4.13

0.79

13.00

a These

a

B

sky types are associated with no sunlight therefore the formula for these cases are not valid

Here A, B, C, D, and E are parameters characterizing a certain sky standard and the values of these parameters are given in Table 2.

3 Daylight Coefficient (DC) Method The daylight illuminance of a room is mainly influenced by the luminance and patterns of the sky in the direction of view of the window at any given time. The Daylight Coefficient [Dγ α ] is defined by the ratio of total illuminance at a particular point to the product of luminance of that sky element and the solid angle subtended by sky element at that point and mathematically expressed as [14, 15]

Comparative Study on Simulation of Daylighting Under …

Dγ α =

E γ α L γ α ∗ Sγ α

187

(5)

where E γ α = illuminance at any station point for the sky element specified by angles γ, α L γ α = luminance of that sky element Sγ α = solid angle subtended by above sky element at the station point. The point-specific illuminance [E] at any horizontal station point due to sky is computed by α2 γ2 E=

L γ α ∗ sinγ ∗ cosγ ∗ dγ ∗ dα

(6)

α1 γ1

4 Simulation: Background Theory In this simulation, Daylight Coefficient (DC) method is applied to compute pointspecific illuminance on horizontal and vertical planes [8, 9]. The total daylight illuminance at a point on working plane within a room is sum of the direct component—the daylight contribution due to the sky portion visible through window opening from that point and the indirect component—the daylight contribution due to interreflection of daylight within the room. Here a room is considered having a single window opening. Variable transmittance of commercially available glazing material as a function of angle of incidence is considered. Room surfaces are assumed as perfect diffuse surface which follows Lambert’s Law of emission. Room details, location, date, month, sky type, etc., are considered as input variables of this developed program. MATLAB coding is used here to take the facilities of its graphical data presentation.

4.1 Grid-Specific Direct Illuminance Computation This is computed for the unobstructed sky as seen from the k-th grid point which is given by Eq. (7). α2 γ2 Edir K

=

L γ α ∗ sinγ ∗ cosγ ∗ dγ ∗ dα α1 γ1

(7)

188

A. Gupta and S. Mukherjee

4.2 Grid-Specific Indirect Illuminance Computation As mentioned earlier, FEM is applied to compute interreflected illuminance. All room surfaces are divided into “m” number of surface elements. The dimension of the surface elements should be sufficiently small compared to the room dimension to have accuracy of computation. All surface elements on floor receive daylight from sky and those on ceiling receive ground reflected daylight. The vertical surface elements on walls receive daylight from sky and ground. According to the theory of Radiosity, the total illuminance on ith surface element can be expressed as [1] E i = E di + ∫ E j ∗ ρ j ∗ F ji

(8)

where E di : direct component or initial illuminance on ith element E j : total illuminance on jth element ρ j : reflectance of jth element F ji : form factor between jth and ith element (from jth to ith) The radiosity is defined as the total rate of energy leaving a surface and equal to the sum of emitted and reflected energy. The reflection modeling is done based on this theory of radiosity and given by  Radiosit yi = Re f electivit yi ∗

Radiosit y j ∗ For m Factor ji

enclosur e

The above concept of reflection modeling can be utilized to compute successive individual reflection where the final illuminance on ith surface element is given by [1] E i = E Oi +

m 

E j .ρ j .F ji

(9)

j=1

Or E i = E Oi +

m 

ai j .E j

j=1

where E Oi : initial illuminance (direct daylight contribution) on ith element

ai j = ρ j .F ji E j : final illuminance on jth element

(10)

Comparative Study on Simulation of Daylighting Under …

189

If E1i be the illuminance on the “ith” surface element from first order reflected light off all other elements then, E11 = 0.EO1 + a12 · EO2 + a13 · EO3 + · · · · · · · · · + a1m EOm E12 = a21 · EO1 + 0.EO2 + a23 · EO3 + · · · · · · · · · + a2m EOm .................................................................. E1m = am1 · EO1 + am2 · EO2 + am3 · EO3 + · · · · · · · · · + 0EOm The above set of equations can be expressed in matrix form as ⎤ ⎡ 0 E11 ⎢ E12 ⎥ ⎢ a21 ⎥ ⎢ ⎢ ⎣ ... ⎦ = ⎣ ... E1m am1 ⎡

a12 0 ... am2

a13 a23 ... am3

⎤⎡ ⎤ E01 . . . . . . . . . . . . . . . a1m ⎢ ⎥ . . . . . . . . . . . . . . . a2m ⎥ ⎥⎢ E02 ⎥ ⎦ ⎣ ..................... ... ⎦ ..................0 E0m

or E1 = A.E0

(11)

In the same way, the illuminance values due to second order interreflection is given by E2 = A.E1

(12)

E2 = A2 .E0

(13)

or,

In general, for the nth order interreflection, the reflected illuminance values are given by En = An .E0

(14)

Hence, the total illuminance matrix E will be the summation of direct component matrix E0 and reflected component matrix, i.e. E = E0 + E1 + E2 + · · · · · · · · · + En

(15)

Equation (15) equation can be rewritten as E = E0 + A.E0 + A2 E0 + · · · · · · · · · + An .EO or

(16)

190

A. Gupta and S. Mukherjee

  E = I + A + A2 · · · · · · · · · + An .E0

(17)

In practice up to fourth-order interreflection is considered since the contribution of reflected light beyond 4th order becomes insignificant and above equation becomes   E = I + A + A2 + A3 + A4 .E0

(18)

Equation (18) gives the approximate total illuminance value on each surface element after fourth-order interreflection.

5 Case Study 5.1 Different Input Parameters In this case study, a room of following specifications is considered for dynamic simulation. • • • • • •

Room dimension: 6 m * 5 m * 3 m. Window dimension: Window area equals to 20% of floor area (30 m2 ) is taken. Wh (window height) = 1.5 m (fixed). Width of window: 4 m. Ceiling, wall and floor reflectance: 60, 60 and 10%. Glazing material: Oceanic Blue.

5.2 Horizontal Angle of Acceptance (α) and Vertical Angle of Acceptance (γ ) Computation of horizontal angle of acceptance (α) with respect to different station points (xs, ys, zs) in mentioned room are considered for five cases of which case 1 is described with suitable diagram (Fig. 2) and vertical angle of acceptance (γ) is shown in Fig. 3. Case 1: When xs < xu and xs < xv, Case 2: When xs = xu and xs < xv, Case 3: When xs > xu and xs < xv, Case 4: When xs = xv and xs > xu and Case 5: When xs > xu and xs > xv Case 1: When xs < xu and xs < xv ∠CAB or ∠BAC = angle1 (say) ∠CBA or ∠ABC = angle2 (say) ∠CAB Calculation: −→ −→ −→−→ AB. AC =  AB  AC Cos∠BAC

Comparative Study on Simulation of Daylighting Under …

191

Fig. 2 Horizontal angle of acceptance

Fig. 3 Vertical angle of acceptance

Cos∠CAB =

−→ −→ −AB. AC   →−→  AB  AC 

Coordinates of C, A and B are C (xs, ys, zs), A (xu, yu, zu), B (xv, yv, zv)

Cos∠CAB =

−→ AB = (xu − xv)iˆ + (yu − yv) jˆ + (zu − zv)kˆ −→ ˆ AC = (xu − xs)iˆ + (yu − ys) + j(zu − zs)kˆ (xu−xv)(xu−xs)+(yu−yv)(yu−ys)+(zu−zv)(zu−zs) 1

[(xu−xv)2 +(yu−yv)2 +(zu−zv)2 ] 2 ∗[(xu−xs)2 +(yu−ys)2 +(zu−zs)2 ]

1/2

angle1 = ∠CAB = cos−1 p alphalft = angle1 − π2 ; In above Fig. 3a αL = alphalft; αR = alphargt alphalower = Wnom + alphalft ∠CBA or ∠ABC Calculation: Coordinates of A, B, and C are

= p(say)

192

A. Gupta and S. Mukherjee

−→ −→ AB. BC Cos∠CAB = −→−→  AB  BC 

A(xu, yu, zu); B(xv, yv, zv); C(xs, ys, zs) −→ AB = (xu − xv)iˆ + (yu − yv) jˆ + (zu − zv)kˆ −→ BC = (xv − xs)iˆ + (yv − ys) jˆ + (zv − zs)kˆ (xv − xu)(xv − xs) + (ys − yv)(yv − ys) + (zu − zv)(zv − zs) 1/2 1  2 [(xu − xv) + (yu − yv)2 + (zu − zv)2 ] 2 ∗ (xv − xs)2 + (yv − ys)2 + (zv − zs)2

= q(say)

angle2 = ∠CBA = cos −1 q alphaupper = Wnom + alphargt Where, Room dimension: L (length), W (width), H (height) Window height: Wh .

5.3 Simulation Results It is identified that sky type 15 (VI.6) prevails in Roorkee during summer and equinox and sky type11 (IV.4) prevails in winter [6]. Keeping the identified sky type fixed, the variation of sky luminance pattern for different seasons are considered after MATLAB simulation with the previous input values. Figure 4a–c represents daylight isolux diagram for summer, equinox, and winter. Figure 4a represents daylight isolux diagram for summer and corresponding gridspecific daylight distribution is given by total_daylight = 95 210 570 949 1362 1543 1543 1362 949 570 209 94 140 242 428 679 957 1168 1168 956 679 427 241 140 152 224 328 468 616 719 719 616 468 328 222 152 145 193 252 328 402 451 451 402 328 252 192 145 129 162 197 237 276 299 299 275 237 196 161 128 115 136 156 180 200 212 211 199 179 155 135 113 101 118 129 142 153 160 159 152 140 127 115 99 93 106 111 118 125 128 128 124 116 108 101 89 90 100 103 107 111 113 112 109 105 99 95 84 83 90 91 94 96 98 97 95 91 88 85 78

Figure 4b represents isolux diagram for the season equinox and corresponding point-specific daylight distribution on working plane is given below

Comparative Study on Simulation of Daylighting Under …

(a)

40 0

600

40 0

3

20 0

Room Depth(m)

3.5

40 0

0 60

4

0 14 00 00 0 0 0 1 80 12

10 00 80 0

20 0

4.5

60 0

Grid-specific daylight illuminance distribution on horizontal working plane : Time-12noon; Day-15 June; North facing window; CIE sky type- 15(VI.6) 5

200

20 0

2.5

200

2 1.5 1 0.5 0

1

0

2

4

3

5

6

Room Length(m)

(b)

Room Depth(m)

10 0

20 0

0 5045 00 0 4

0 20

3

15 0

15 0

2.5

100

300 25 0

0 15 10 0

3.5

550

50 0 45 0 40 0 35 0

150

4

0 0 25 30

20 0

4.5

35 0 30 25 0 0 20 0

Grid-specific daylight illuminance distribution on horizontal working plane : Time-12noon; Day-15 March; North facing window; CIE sky type- 15(VI.6) 5

0 10

10 0

2

10 0

1.5 1 0.5 0

6

5

4

3

2

1

0

Room Length(m)

(c)

25 0

0 30 25 0

0 20

15 0

2.5

4 4 50 350 0 0

15 0 0 15

20 0

450 400 35 0

0 15

30 0

00 2250

60 0

35 0

3.5 3

0 65 60 0 0 55 0 0 5

50 606 0 55 0 50 0

0 0 0 45 4

25 0

4

30 0

4.5

200

Grid-specific daylight illuminance distribution on horizontal working plane : Time-12noon; Day-15 January; North facing window; CIE sky type-11(IV.4) 5

Room Depth(m)

Fig. 4 a Daylight isolux diagram: summer (June), north facing window, 12 noon. b Daylight isolux diagram: equinox (March), north facing window, 12 noon. c Daylight isolux diagram: winter (January), north facing window, 12 noon

193

200 2 0 15

15 0

1.5

15 0

1

0 15

15 0

0.5 0

0

1

2

3

Room Length(m)

4

5

6

194

A. Gupta and S. Mukherjee total_daylight = 111 183 420 597 696 560 560 694 594 416 180 108 151 219 333 462 562 601 600 560 460 331 218 149 165 214 274 354 422 463 463 422 352 273 213 164 162 196 228 274 316 340 340 315 273 228 195 163 150 175 191 216 241 254 254 240 216 191 173 148 142 158 165 178 191 200 200 190 177 163 157 139 132 147 149 155 162 166 165 160 153 145 142 127 132 145 143 145 148 150 149 146 141 137 136 124 139 152 149 148 150 151 151 147 144 140 140 128 44 141 139 141 142 141 139 135 132 133 127 138

Figure 4c represents isolux diagram for winter season and corresponding pointspecific daylight distribution on working plane is given below Daylit zone

Partially Daylit zone

Non Daylit zone

total_daylight = 71 134 304 460 588 564 563 582 451 294 125 66 97 149 234 336 432 490 487 426 328 225 142 92 102 138 184 243 298 333 331 294 238 178 132 97 97 121 146 179 209 227 226 207 176 142 116 93 85 103 118 135 152 162 161 151 133 115 100 83 77 89 97 107 116 122 121 115 105 95 87 75 69 79 83 88 93 96 96 93 87 80 75 66 66 73 75 78 80 81 81 78 75 71 69 61 66 72 72 73 74 75 75 73 71 68 67 60 62 65 65 66 67 68 67 65 64 62 61 57

6 Results and Analysis In the present work, daylight prediction database is generated in terms of average illuminance and overall uniformity on entire working plane for eight different window orientations and five different local times for three different seasons and the corresponding bar diagram plots are presented in Fig. 5a–c. In the integrated lighting simulation, the contributions from both the daylighting system and from the artificial lighting system are simultaneously taken into account which results energy savings [16, 17].

Comparative Study on Simulation of Daylighting Under …

(a) Average E(Lux) in Daylight area

Fig. 5 a Average illuminance (Lux) in daylight area for eight window orientations J = 166 (summer). b Average illuminance (Lux) in daylight area J = 74 (equinox). c Average illuminance (Lux) in daylight area J = 15 (winter)

195

1400 1200

8am

1000

10am

800 600

12noon

400

2pm

200

4pm

0 N

NE

(b) Average E(Lux)in Daylight area

E

SE

S

SW W NW

Window orientaƟon

1200 1000

8am

800

10am

600

12noon

400

2pm

200

4pm

0 N

NE

E

SE

S

SW

W NW

Window orientaƟon

Average E(Lux) in daylit area

(c) 1000 8am

800

10am

600

12noon 400

2pm

200

4pm

0 N

NE

E

SE

S

SW W NW

Window orientaƟon

7 Conclusion Simulation results for three seasons, different window orientations considering the whole day, show that the illuminance value on the daylight area varies in different manner depending on window orientations. In case of south and north window orientation the above-mentioned parameter almost remains constant. But in case of east and west window orientations these values vary on the basis of time and season. Considering the nature of data variation the position of the windows of any build-

196

A. Gupta and S. Mukherjee

ing in Roorkee can be designed so that maximum daylight can penetrate into the room through window. On the whole, in amount of daylight penetration is maximum compare to equinox and winter solstice. This is due to the earth’s polar axis is tilted 23.5° to the orbital plane (ecliptic plane). Combinations of rotation, revolution, and tilt of the polar axis result in differential illumination and changing illumination patterns on Earth. During summer sun shines directly on the northern hemisphere and indirectly on the southern hemisphere. During winter sun shines directly on the southern hemisphere and indirectly on the northern hemisphere. During equinox sun shines equally on the southern and northern hemisphere. Acknowledgements The author wishes to acknowledge the support received from Dr. R. Kittler and Dr. Danny H. W. Li since they sent their publications which helped to complete this research. She also likes to thank Indian Society of Lighting Engineers [ISLE] and Mr. P. K. Bandyopadhyay, past President, ISLE for providing a copy of the report [12] published by Central Building Research Institute [CBRI] containing Indian Measured Daylight Database.

References 1. Tregenza PR, Waters IM (1983) Daylight coefficient. Light Res Technol 15(2):65–71 2. Li DHW, Lau CCS, Lam JC (2011) A simplified procedure using daylight coefficient method for sky component prediction. Arch Sci Rev 14(3):287–294 3. Investigations on evaluation of daylight and solar irradiance parameters for improved daylighting of buildings and energy conservation in different climates. Central Building Research Institute [CBRI] (2001) 4. Li DHW, Lau CCS, Lam JC (2003) A study of 15 sky luminance patterns against Hong Kong Data. Arch Sci Rev 46(16):1–68 5. Kittler R, Darula S (2006) The method of aperture meridians: a simple calculation tool for applying the ISO/CIE standard general sky. Light Res Technol 38(2):109–122 6. Mukherjee Sutapa, Roy Biswanath (2012) Correlating Indian measured sky luminance distribution and Indian Design clear sky model with five CIE standard clear sky models. J Opt 40(4):150–161 7. Darula S, Kittler R, CIE general sky standard defining luminance distributions. Institute of Construction and Architecture, Slovak Academy of Sciences 9, Dubravska Road, SK—842 20 Bratislava, Slovakia 8. Reinhart Christoph F, Herkel S (2000) The simulation of annual daylight illuminance distributions—a state-of-the-art comparison of six RADIANCE-based methods. Energy Build 32(2):167–187 9. Mardaljevic J (2000) Simulation of annual daylighting profiles for internal illuminance. Light Res Technol 32(3):111–118 10. Li DHW, Tang HL (2008) Standard skies classification in Hong Kong. J Atmos Sol Terr Phys 70(8):1222–1230 11. CIE—Commission Internationale de l Éclairage (2003) Spatial distribution of daylight-CIE Standard General Sky. CIE Standard S 011/E. CIE Central Bureau, Vienna 12. Wright J, Perez R, Michalsky J (1989) Luminous efficacy of direct irradiance: variations with insolation and moisture conditions. Sol Energy 42(5):387–394 13. Debashis Raul, Sujoy Pal, Biswanath Roy (2015) Application of Perez daylight efficacy model for Kolkata. J Inst Eng India Ser B 96(4):339–348 14. Littlefair PJ (1992) Daylight coefficients for practical computation of internal illuminances. Light Res Technol 24(3):261–266

Comparative Study on Simulation of Daylighting Under …

197

15. Li DHW, Cheung GHW, Lau CCS (2006) A simplified procedure for determining indoor daylight illuminance using daylight coefficient concept. Build Environ 41(5):578–589 16. Li DHW (2007) Daylight and energy implications for CIE standard skies. Energy Convers Manag 48(3):745–755 17. Fernandes LL, Lee ES, Ward G (2013) Lighting energy savings potential of split-pane electrochromic windows controlled for daylighting with visual comfort. Energy Build 61C:8–20

Application of Modified Harmony Search and Differential Evolution Optimization Techniques in Economic Load Dispatch Tanmoy Mulo, Prasid Syam and Amalendu Bikash Choudhury

Abstract Present day civilization faces a never-ending growth for the demand of electricity. This necessitates an increase in the number of power stations and their capacities and consequent increase in the power transmission network connecting the generating station to load centers. Depending upon the load demand, the electrical generators operate under various generating conditions. The generating costs of different power plants are also different. So it is very much important to operate the power plant at optimal generation with minimum cost condition. In this paper, an attempt has been made in minimizing the cost function and the transmission line losses utilizing Modified Harmony Memory Search (MHMS), and Differential Evolution (DE) optimization techniques for a three-unit system under known maximum and minimum operating region of each generating station. Keywords Harmony search (HS) · Modified harmony memory search (MHMS) · Differential evolution (DE) · Economic load dispatch (ELD)

1 Introduction A practical power system network comprises a large number of buses interconnected by long transmission lines. Active and Reactive Power is injected into a bus from generators. Throughout the past decade, generally, GA [1] method is preferred in optimizing the economic load dispatch [2] problem, but recent research has pointed few deficiencies in GA optimization [3] method like time to reach the best feasible solution, complexity with the increasing number of Buses, etc. The problem T. Mulo (B) · P. Syam · A. B. Choudhury Electrical Engineering Department, Indian Institute of Engineering Science and Technology, Shibpur, Shibpur, India e-mail: [email protected] P. Syam e-mail: [email protected] A. B. Choudhury e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_16

199

200

T. Mulo et al.

associated with the large execution time can be solved by using parallel [1] computing platform that is based on a network of interconnected personal computers (PC) using TCP/IP socket communication facilities. Optimization technique is being used nowadays in various places for maximization or minimization of any function in real time by using DE [4], PSO [5, 6], GA techniques. The proposed Modified Harmony Memory [7] Search method considers the nonlinear behavior of generator cost function to minimize transmission losses as well as operating cost simultaneously for multi-objective purpose [8–10]. In general, DE is known to be the fastest algorithm among all optimization techniques but for less values of initial population it converges apart from its global position [11–13]. However, the proposed Modified Harmony Memory Search (MHMS) algorithm is less dependent on initial memory value and also converges near to global value in less CPU time as compared to DE. Case I We consider a three bus and three generating unit system without any losses in transmission lines as shown in Fig. 1. This implies that the B matrix components are all equal to zero. The objective is to minimize the cost function [14] given in Eq. (1) [15] using Differential Evolution and Modified Harmony Search optimization techniques (Fig. 2). F1 = 0.00525P1 2 + 8.663P1 + 328.13 Rs/h F2 = 0.00609P2 2 + 10.040P2 + 136.1 Rs/h F3 = 0.00592P3 2 + 9.76P3 + 59.16 Rs/h where 50 MW ≤ P1 ≤ 250 MW 5 MW ≤ P2 ≤ 150 MW 15 MW ≤ P3 ≤ 100 MW Min (F1 , F2 , F3 ) = Min (summation of F1 , F2 , F3 )

Fig. 1 Three-bus system with single load

(1)

Application of Modified Harmony Search … Fig. 2 Flow chart of differential evolution algorithm

201

202

T. Mulo et al.

1.1 Initialization First of all the generating random set of data in between a given range (Maximum and minimum) using Eq. (1) P1 = 50 + rand (1, 1) ∗ (200); P2 = 5 + rand (1, 1) ∗ (145); P3 = 300 − P1 − P2 ;

(2)

1.2 Mutation Using two values in the initial population formation of new values as formula given Eq. (3) ViiG = Xmin + F∗ (P2i − P1i )

(3)

The updated new value with scale factor F is a fixed value in between (0, 2).

1.3 Recombination Components of the donor vector enter into the trial offspring vector in the following way: Let jrand be a randomly chosen integer between 1, …, D. Binomial (Uniform) Crossover   V elo j,i,G , i f randi, j [0, 1] ≤ Cr or j = jrand U 1i,G = (4) a j,i,G , other wise

1.4 Selection From the principle of “Survival of the fitter” The trial offspring vector is compared with the target (parent) vector and the one with a better fitness is admitted to the next generation population as per Eq. (5).  ai,G+1 =

    U 1i,G , i f f U 1i,G ≤  f ai,G ai,G , i f f U 1i,G > f ai,G

(5)

Application of Modified Harmony Search …

203

Result of Differential Evolutionary Optimization: Table 1a–c are the data for number of population 10,50, 1000 respectively conTable 1 DE method data Maximum iterations

P1

P2

Fitness value

Time (s)

P3

10

177.8

57.6

3483.4

0.45185

64.6

50

184.3

45.4

3482.1

0.48077

70.3

100

182.4

46.8

3482.1

0.47330

70.8

150

183.9

45.5

3482.1

0.51906

70.6

200

183.9

45.5

3482.1

0.51745

70.6

250

183.8

45.5

3482.1

0.48248

70.7

300

184.0

45.4

3482.1

0.48142

70.6

350

174.2

48.3

3482.9

0.48232

70.4

400

186.8

49.5

3482.5

0.47858

63.7

450

184.5

44.6

3482.1

0.48582

70.9

10

186.12

44.45

3482.1

0.46082

69.43

50

184.16

45.37

3482.1

0.50693

70.47

100

183.92

45.60

3482.1

0.56211

70.48

150

183.20

46.68

3482.1

0.58283

70.12

200

183.97

45.53

3482.1

0.65434

70.5

250

183.96

45.53

3482.1

0.75007

70.51

300

183.97

45.53

3482.1

0.80025

70.5

350

183.96

45.53

3482.1

0.69671

70.51

400

183.96

45.53

3482.1

0.65218

70.51

450

183.96

45.53

3482.1

0.67221

70.51

10

184.46

45.42

3482.1

0.60496

70.12

50

184.07

45.48

3482.1

1.38608

70.45

100

183.87

45.58

3482.1

1.55524

70.55

150

183.96

45.53

3482.1

3.53172

70.51

200

183.96

45.53

3482.1

4.55105

70.51

250

183.96

45.53

3482.1

5.60395

70.51

300

183.96

45.53

3482.1

4.25355

70.51

350

183.96

45.53

3482.1

4.81914

70.51

400

183.96

45.53

3482.1

7.19371

70.51

450

183.96

45.53

3482.1

8.11083

70.51

A

B

C

204

T. Mulo et al.

sidering same scale factor 0.7, Probability 0.5 (Data Mutation) in case of DE optimization (Figs. 3, 4, 5 and 6). Results of Modified Harmony Memory Search Algorithm Optimization In the proposed MHMS method new selection of data methodology is being replaced as compared to HMS method. In HMS method the randomly data selection is 10%, data consideration from hms matrix is 81 and 9% of data are sorted by delta shift to hms data. However, in the proposed Modified Harmony Memory Search (MHMS) method, this data selection is made as mentioned below. Harmony Memory Considering Rate (HMRC) = 0.9; Pitch Adjusting Rate (PAR) = 0.1; In this method the data can be selected by HMCR and PAR HMCR * PAR = 0.09 = 9% data with delta shift (0.05–0.1) of first value 81% = (1 − PAR) * HMRC data randomly generated and checking it position in memory by appropriate rest of data selected first value. Three cost functions are to be minimized as given in Eq (1)

Fig. 3 Fitness value versus variable P1 and P2 for 10 populations

Fig. 4 Fitness value versus variable P1 and P2 for 50 populations

Application of Modified Harmony Search …

205

Fig. 5 Fitness value versus variable P1 and P2 for 1000 populations

Cost_function = f(P1 , P2 ) = (0.00525P21 + 0.00609P22 + 0.0 0592(300(P1 + P2 ))2 + 8.663P1 + 10.040P2 + 9.76 * (300 − (P1 + P2 )) + 524.2); P1 = 50 + rand (1, 1) * (200); P2 = 5 + rand (1, 1) * (145); P3 = 300 − P1 − P2 ; (as the total generation required 300 MW neglecting losses in transmission line [15]) (Tables 2 and 3). From Table 4a it has been observed that Modified Harmony Memory Search is faster than DE, although the value does not reaches exactly to the optimal placement but still it is faster. In case of DE, same ten set of data for 100 iterations to reaches the optimal solution take 0.33825 s whereas in case of MHMS takes 0.21127 s. That is why further investigation of ELD problem is made utilizing Modified Harmony Memory Search with little increment of Functional evaluation value such that it reaches very near to optimal value with in short time. DE is fastest if the number of initial value or initial population reduces; however, it will converge somewhere apart from the global position. This is one of the major disadvantages of DE algorithm. Harmony search, on the other hand always converge to either Global or somewhere near to global as per the requirement. Tables 4, 5, 6 and 7 are the Data updated in MHMS Memory after 100, 1000, 10000 &100000 numbers of functional evaluations in 0.288124 s, 0.293496 s, 0.498758 s & 2.675813 s respectively (Fig. 7). Case II: All the above calculations have been completed considering losses equal to zero or neglecting line losses. In Case II, we consider the loss parameter B matrix and the generator can supply 286.081 MW of load demand as shown in Fig. 8. For example,

206

T. Mulo et al.

Fig. 6 Flow chart of modified harmony memory search algorithm



⎤ 0.000136 0.000175 0.000184 B = ⎣ 0.000175 0.000154 0.000283 ⎦ 0.000184 0.000283 0.000161 Then for this optimized value the loss can be calculated with the help of Eq. (6). In this case P1 = 184 MW, P2 = 46 MW, P3 = 70 MW.

P = 184 46 70 PL =

m m i=1 j=1

Pi Bi j P j +

m i=1

B0i Pi + B00

Application of Modified Harmony Search … Table 2 No of population 10 or initial memory size, random data generated in MHMS memory at the begin

207

P1

P2

238.7

115.2

3619.8

227.915

50.70121

3507.448

151.5394

65.85238

3491.77

175.3636

43.05408

3484.022

124.4949

29.56142

3536.691

152.9587

127.5766

3544.319

223.7209

125.3921

3614.69

154.0259 132.3622

Table 3 Ranking given on the basic of fitness value

Fitness_value

21.61113

3508.24

54.72312

3508.015

203.1605

113.0801

3557.12

109.4852

149.8776

3583.57

P1

P2

Fitness_value

43.05408

3484.022

1

151.5394

65.85238

3491.77

2

227.915

50.70121

3507.448

3

132.3622

54.72312

3508.015

4

154.0259

21.61113

3508.24

5

124.4949

29.56142

3536.691

6

152.9587

127.5766

3544.319

7

203.1605

113.0801

3557.12

8

109.4852

149.8776

3583.57

9

223.7209

125.3921

3614.69

10

⎡ ⎤⎡ ⎤ 184 0.000136 0.000175 0.000184 PL = 184 46 70 ⎣ 0.000175 0.000154 0.000283 ⎦⎣ 46 ⎦ 0.000184 0.000283 0.000161 70

Rank

175.3636

(6)

B0i = 0and B00 = 0, Loss PL = 13.9109 MW at optimal generation in minimum operating cost. This three bus system can provide = (300 − 13.9109) MW = 286.0891 MW load demand. Case III: In Case II, it is considered that the optimal value is fixed and accordingly the load demand is reduced by loss amount. In this case, we consider the B matrix parameters are fixed for a particular three-bus system. The generation unit supplies an optimal set of value to minimize losses and minimize cost of generation simultaneously by without violating Eq. (7).

208

T. Mulo et al.

Table 4 Harmoy Memory after iteration 100 P1

P2

Fitness_value

187.5779957

54.61523003

3484.390904

Rank 1

167.1004428

54.08885081

3485.215923

2

167.0954428

54.08385081

3485.2172736

3

170.5255047

40.7607982

3485.9203170

4

170.5205047

40.75579826

3485.9234717

5

164.8040141

53.39289413

3485.9284272

6

162.4220540

57.32742516

3486.7145807

7

206.5100966

38.03042552

3487.2171577

8

147.9144639

67.62213689

3493.8168844

9

220.4263598

20.66332208

3494.409043

10

A Time comparison of two methods MHMS method

P1

165.51

179.08

186.44

184.74

P2

60.87

46.74

45.51

44.82

45.67

FV

3486.1

3483.1

3482.9

3482.9

3482.9

P1

199.97

183.96

183.96

183.96

183.96

P2

38.52

45.47

45.53

45.53

45.53

FV

3485.9

3482.9

3482.9

3482.9

3482.9

No of iterations/functional evaluations

10

100

1000

10000

100,000

MHMS time (s)

0.21124

0.21127

0.24536

0.54207

3.56691

DE time (s)

0.30542

0.33825

0.59113

3.13362

28.57982

DE method

183.85

Initial population = 10. Time comparison of two methods as shown above is simulated in AMD A6 4 GB RAM 64-bit Operating System using MATLAB 2012b Table 5 Harmoy Memory after iteration 1000

P1

P2

Fitness_value

Rank

184.8318949

42.961319245

3482.92940928

1

182.262016480

49.4654340946

3483.00610778

2

179.812734781

45.6958937063

3483.05302098

3

178.913899298

48.8979986808

3483.08747436

4

184.902946739

49.6792034737

3483.12929035

5

182.90532859

41.1691265088

3483.16447363

6

177.825177950

50.9140458586

3483.24521415

7

179.666882829

52.1062551409

3483.25793445

8

177.971270339

53.2801387581

3483.43949559

9

176.961675861

44.0582824864

3483.56494253

10

Application of Modified Harmony Search … Table 6 Harmoy Memory after iteration 10000

Table 7 Harmoy Memory after iteration 100000

209

P1

P2

Fitness_value

Rank

183.86122895

43.841786955

3482.9045056

1

185.72799168

46.018067283

3482.9150876

2

186.46571106

44.984961010

3482.9247262

3

184.19624995

47.603348825

3482.9250933

4

184.20124995

47.608348825

3482.9255036

5

184.20624995

47.613348825

3482.9259157

6

182.10053901

48.102094481

3482.9288907

7

181.6541964

45.466758281

3482.9294660

8

184.2010775

47.71997540

3482.931507

9

182.14267756

48.566967415

3482.949614

10

Rank

P1

P2

Fitness_value

184.08180993

45.451817185

3482.86780691

1

184.08680993

45.456817185

3482.86781188

2

184.08746175

45.682197449

3482.86830322

3

184.47033437

45.438148239

3482.87003944

4

183.59298445

46.209027120

3482.87168400

5

183.32606486

45.463885753

3482.87291005

6

183.35179327

45.315158139

3482.87414124

7

183.04128262

45.627243191

3482.87638349

8

184.98795903

44.903963536

3482.8764922

9

185.06566238

44.920820903

3482.87771388

10

Fig. 7 Fitness value versus variable P1 and P2 for MHMS

210

T. Mulo et al.

Fig. 8 Three bus systems with one single load connection 3 i=1

Algorithm III See Fig. 9.

Fig. 9 Flow chart of MHMS algorithm

PG i − PDi − PL i = 0.

(7)

Application of Modified Harmony Search …

211

Procedure Step 1: At first, a randomly set of value is generated as per HMCR and PAR. Step 2: Then for each set of values, the cost function value is estimated. Step 3: Then we are sorting the data in accordance with cost value ascending. Step 4: Calculate the loss value as P * B * P . Step 5: Check the total generation. (Tg) ~ (losses + load demand) as stopping criteria. Both the cost function and losses are minimum at 4th column as shown in Table 8. RANKSUM method is being used for multiobjective purpose as shown in Fig. 10. The sum of normalized data is minimum at the point where both the cost function and losses are minimum. P1 = 203.8956 MW, P2 = 90.7288 MW, P3 = 18.2101 MW. Table 8 MHMS technique for optimal generation, population size and time with fixed load demand Td = 300 MW P1

211.6

80.8

182.2

203.8

239.1

P2

81.7

142.2

P3

19.5

95.04

No of population (×103 )

20

Tg Td

57.8

90.7

28.4

60.40

75.34

18.21

46.84

36.48

35

55

85

90

100

312.9

318.1

315.4

312.8

314.4

313.8

300

300

300

300

300

300

Loss

12.940

18.15

15.50

12.83

14.44

13.86

Cost value (Rs./h)

3650.0

3790.

3650.0

3650.0

3660

3640.0

Time (s)

23.94

81.10

205.73

498.05

612.8

729.0

Fig. 10 Normalized value of cost value and the losses values

216.96

212

T. Mulo et al.

2 Conclusion This paper deals with the optimization technique in Economical load Dispatch using Harmony and Differential Evolution Optimization technique to generate minimum cost and optimal operating value of three generators supplying a load. From analysis, it may be concluded that Modified Harmony Memory Search gives same result in lesser time for more number of population size and fitness value estimation as compared to DE. Considering the B matrix, the load demand can be fulfilled by modified Harmony Memory Search technique as discuss in algorithm III and applying RANKSUM method where minimum operating cost and line losses are both achieved simultaneously.

References 1. Fung CC, Chow SY, Wong KP (2000) Solving the economic dispatch problem with an integrated parallel genetic algorithm. 0-7803-6338-8/00/$10.00(~)2000, IEEE 2. Santos MRBD, Balbo AR, Gonçalves E, Soler EM, Pinheiro RBNM, Nepomuceno L, Baptista EC (2017) A proposed methodology involving progressive bounded constraints and interiorexterior methods in smoothed economic/environmental dispatch problems. IEEE Lat Am Trans 15(8) 3. Hui S, Suganthan PN (2015) Ensemble and arithmetic recombination-based speciation differential evolution for multimodal optimization. IEEE Trans Cybern (Online since Mar 2015). https://doi.org/10.1109/tcyb.2015.2394466 4. Das S, Mullick SS, Suganthan PN (2016) Recent advances in differential evolution—an updated survey. Swarm Evol Comput 27:1–30 5. Yoshida H, Kawata K, Fukuyama Y, Takayama S, Nakanishi Y (2000) A particle swarm optimization for reactive power and voltage control considering voltage security assessment. IEEE Trans Power Syst 15:1232–1239 6. Zhao SZ, Suganthan PN (2011) Two-lbests based multi-objective particle swarm optimizer. Eng Optim 43(1):1–17 7. Geem ZW, Kim JH, Loganathan GV (2001) A new heuristic optimization algorithm: harmony search. Simulation 78(2):60–68 8. Qu BY, Suganthan PN (2010) Multi-objective evolutionary algorithms based on the summation of normalized objectives and diversified selection. Inf Sci 180(17):3170–3181 9. Qu BY, Suganthan PN (2011) Multi-objective differential evolution based on the summation of normalized objectives and improved selection method. In: SDE-2011 IEEE symposium on differential evolution, Paris, France 10. Zhou A, Qu B-Y, Li H, Zhao S-Z, Suganthan PN, Zhang Q (2011) Multiobjective evolutionary algorithms: a survey of the state-of-the-art. Swarm Evol Comput 1(1):32–49 11. Qu BY, Suganthan PN (2009) Multi-objective evolutionary programming without nondomination sorting is up to twenty times faster. IEEE congress on evolutionary computation, pp 2934–2939, Norway 12. Qu BY, Suganthan PN (2011) Multi-objective differential evolution based on the summation of normalized objectives and improved selection method. SDE-2011 13. Zeng X, Liu Z, Hui Q (2015) IEEE energy equipartition stabilization and cascading resilience optimization for geospatially distributed cyber-physical network systems. IEEE Trans Syst Man Cybern Syst 45(1)

Application of Modified Harmony Search …

213

14. Subathra MSP, Easter Selvan S, Aruldoss Albert Victoire T, Hepzibah Christinal A, Amato U (2015) A hybrid with cross-entropy method and sequential quadratic programming to solve economic load dispatch problem. IEEE Syst J 9(3) 15. Kothari DP, Dhillon JS. Power system optimization

Design of a Multilevel Inverter Using SPWM Technique Arka Ray, Shuvadeep Datta, Amitava Biswas and Jitendra Nath Bera

Abstract This paper proposes and examines a sinusoidal pulse width modulation (SPWM)-based single-phase diode clamped multilevel inverter for generation of multilevel output voltage. The SPWM signals and digital square pulses are generated using a single PWM module and I/O port of a PIC microcontroller. The performance of the proposed method is checked through simulation after the design of a diode clamp multilevel inverter for 3-level, 5-level, and 9-level output voltage. The 5-level and 3-level output have also been produced by hardware implementation of the designed circuit. Total harmonic distortion (THD) for 3-level, 5-level, and 9-level output voltage waveforms are analyzed. Keywords Diode clamped multilevel inverter · SPWM signal · PIC microcontroller

1 Introduction Multilevel inverter is very essential for medium voltage high power applications such as power industry, tractions, reactive power compensation (STATCOM, UPFC, UPQC, DVR), and also in some low-voltage applications such as in aerospace and class D digital audio power amplifier [1–5].

A. Ray (B) · S. Datta · A. Biswas · J. N. Bera Department of Applied Physics, University of Calcutta, Kolkata, India e-mail: [email protected] S. Datta e-mail: [email protected] A. Biswas e-mail: [email protected] J. N. Bera e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 T. K. Basu et al. (eds.), Advances in Control, Signal Processing and Energy Systems, Lecture Notes in Electrical Engineering 591, https://doi.org/10.1007/978-981-32-9346-5_17

215

216

A. Ray et al.

This inverter usually produces an output voltage either 0 or ±Vdc which is called two-level inverter. In high power, high-voltage applications this two-level inverter has some constraints like-THD, high switching losses, and device ratings [6]. With the use of multilevel concept, voltage levels of the inverter can be increased without using higher ratings of the individual devices. The structure of multilevel voltage source inverter allows it to reach high voltages with low-harmonic contents without the use of a transformer. So avoidance of transformer is possible to reach a higher output voltage. The weight of the overall system is significantly decreased for achieving higher voltage levels. The harmonic content decreases as the level of output voltage increases. The output voltage levels can be extended beyond the levels of individual power electronic device. So for the design of multilevel inverter, it is possible to reduce the EMI problems by reducing the switching dv/dt [7]. For multilevel inverter, the structure helps to reduce the harmonic content of the output voltage waveform so switching frequency of the power electronic devices can be reduced than the traditional 2-level inverter. Due to this, device switching losses become less and the overall efficiency is improved. Modulation methods used in multilevel inverter can be classified according to its switching frequency [8]. Several modulation and control strategies have been used, including sinusoidal pulse width modulation (SPWM), selective harmonic elimination modulation (SHEM), and space vector modulation (SVM). But still, a popular method for the design of multilevel inverter in medium voltage high power application is the carrier-based SPWM [1]. The main reason for selecting SPWM is the modulation method that requires only the reference and carrier signal, and a simple comparator to produce the gating signals. But for generation of “m” level output voltage, “m − 1” carrier signals are required. So getting different output voltage levels of a multilevel inverter, microcontroller should require many independent PWM module, which is practically impossible to find. So a SPWM signal is introduced herewith I/O operation of a microcontroller for generation of multilevel output voltage. This SPWM signal requires only a single PWM module to get any level of inverter output voltage. Sometimes application wise it may require to convert the higher level output voltage to lower level voltage using the same higher level inverter circuit. That can be possible using the introduced SPWM signal.

2 Diode Clamp Multilevel Inverter 2.1 Operating Principle For generation of m-level output voltage of a diode clamp multilevel inverter, the required DC sources are (m − 1). If the input DC bus voltage is Vdc , the voltage across each input DC source is Vdc /(m − 1), and each device voltage stress is one input DC source voltage level, i.e., Vdc /m − 1, through the clamping diodes [9, 10].

Design of a Multilevel Inverter Using SPWM Technique

217

Though we have designed the circuit on the basis of “m” level (for 0 ≤ t ≤ 0.005 s) phase voltage but the output voltage level should be (2m − 1). If we will design a 5-level output (line) voltage, then “m” can be taken as 3. So the number of input DC sources = (m − 1) = (3 − 1) = 2. The number of clamping diodes in the leg of each phase = (m − 1)* (m − 2) = (3 − 1) * (3 − 2) = 2 (Fig. 1). Voltage stress of switching device = Vdc /(m − 1) = Vdc /2. If we will design a 9-level output (line) voltage, then “m” can be taken as 5. So the number of input DC sources = (m − 1) = 4. The number of clamping diodes in the leg of each phase = (m − 1) * (m − 2) = (5 − 1) * (5 − 2) = 12. Voltage stress of switching device = Vdc /(5 − 1) = Vdc /4. Table 1 refers the switching state of a 5-level output (line) voltage.

Fig. 1 A five-level diode clamped multilevel inverter circuit diagram

Sb1

Sa1

Sa2 Db1

Vdc/2 Da1

Load Sb'1

Sa'1

Da2

Db2

Vdc/2 Sb'2

Sa'2

Table 1 Diode clamp 5-level inverter output (line) voltage levels and their switch states Output voltage

Switch state Sa1

Sa2

Sa 1

Sa 2

0

0

0

1

1

Vdc /2

0

1

1

0

Vdc

1

1

0

0

218

A. Ray et al.

Fig. 2 Line voltage waveform of a 5-level inverter

(1) To generate an output voltage of Vdc , the upper leg switches of phase “a” from Sa1 to Sa2 to be turned on. (2) To generate an output voltage of Vdc /2, the upper leg switch Sa2 and a lower leg switch Sa 1 of phase “a” to be turned on . (3) To generate an output voltage of 0, the lower leg switch Sa 1 and Sa 2 of phase “a” to be turned on. Table 1 represents the voltage levels and their respective switching states for a 5level diode clamped multilevel inverter. State condition “1” means that the switches are on, and “0” means the switches are off. In 5-level inverter circuit there is two complimentary switch pairs in each phase. When switch Sa1 is working, switch Sa 1 never works and when switch Sb1 is working, switch Sb 1 never works. Using phase-leg of “a” the example shows the two complementary pairs are (Sa1 , Sa 1 ), (Sa2 , Sa 2 ). For generation of 9-level output voltage, there should be four complimentary switch pairs like (Sa1 , Sa 1 ), (Sa2 , Sa 2 ), (Sa3 , Sa 3 ), (Sa4 , Sa 4 ). Figure 2 shows phase and line voltage waveforms of a 5-level inverter. The line voltage is the resultant voltage of positive phase-leg “ a” and the negative phase-leg “ b”. This implies that the m-level converter has an m-level output phase voltage and a (2m − 1) level output line voltage.

2.2 Advantages • The number of levels in the output voltage is high enough which reduces the harmonic distortion. So the use of filter can be avoided. • The device can be switched in fundamental frequency to improve the efficiency. • Overall controlling action is very simple.

Design of a Multilevel Inverter Using SPWM Technique

219

3 Modulation Techniques for Multilevel Inverter

MODULATION METHOD

CLASSIC

NEW

SVM SPWM

SHEM

Many modulation techniques have been developed to generate the triggering pulses for multilevel inverter switches. The modulation scheme which is usually used in industry is sinusoidal pulse width modulation (SPWM). Though there are some other modulation methods like space vector modulation and selective harmonic elimination modulation which have several advantages like switching frequency reduction, common-mode elimination or reduction, lower THD but still the best probable reason for selecting carrier-based SPWM is that it only needs the modulating and carrier signal, and simple comparator to deliver the gating signal. So the generation of SPWM signal is less tedious than conventional space vector modulation or selective harmonic elimination modulation. But to generate a multilevel output voltage of “m” level using SPWM, we usually require “m − 1” carrier signals which can be level shifted or phase shifted [8]. Now for the implementation of the level shifted or phase shifted SPWM, we require a microcontroller which has “m − 1” independent PWM module. We have used a peripheral interface controller (PIC) for generation of a multilevel output voltage. In this paper, we are introducing a switching window signal for optimization of individual PWM module of a PIC microcontroller. We analyzed the duty cycle of the individual power electronic switches and generate the PWM switching signal using PIC microcontroller’s I/O pin. The conventional unipolar SPWM signal is also generated using the PIC microcontroller at the same instant with a single PWM module. The switching I/O signals and an SPWM signal are passed through a logic gate to generate required SPWM signal for triggering of multilevel inverter switches. This method can optimize the use of individual PWM module of a PIC microcontroller to generate the SPWM signal for triggering pulse of multilevel inverter switches (Fig. 3).

220

A. Ray et al.

Sa1

Duty Cycle = 16.67 %

Sa2

Duty Cycle = 33.33 %

Sb1

Duty Cycle = 16.67 %

Sb2

Duty Cycle = 33.33 %

Fig. 3 Switching sequences

4 Schematic Representation a. Control Signal

b. Multilevel Inverter Diode Battery Unit MLI Switches

Control Signal

Filter

Load (1-ph)

Design of a Multilevel Inverter Using SPWM Technique

221

5 Algorithm for Generation of SPWM Signal The program for generation of 5-level output voltage has been written in assembly language. The steps which have been followed are mentioned below: (1)

Load a counter with a value of m/6 related to the number of samples of lookup table, contained ‘m’ number of duty cycle for half time period of reference signal. (2) Enable a single PWM module in upcounter mode using PWM Timer and the I/O pins of microcontroller for generation of a PWM digital square pulse for a switch state. (3) Read the lookup table for a table index value of duty cycle. (4) Decrement the counter and repeat Step 3. (5) Go to Step 1 when the counter becomes zero. (6) Enable the I/O pins in such a manner to get another switch state. (7) Repeat Step 3 and 4. (8) Repeat Step 5–7 until “m” table index value is read. (9) Repeat the Steps 1–6 for next half time period of reference signal. (10) For generation of 3-level inverter enable I/O pins in a manner different from the 5-level inverter to get different switch states.

6 Experimental Result (a) Required Simulation Software: Software used for simulation and programing • MATLAB • MPLAB IDE • PROTEUS Design Suite (b) Used Hardware Components: Hardware implementation has also been done for generation of 3-level and 5-level output voltage by • • • •

PIC microcontroller(Microchip) IRF MOSFET Clamping Diode Opto-isolator

222

A. Ray et al.

6.1 Simulation Results 6.1.1

Results Found from MATLAB

See Figs. 4, 5, 6, 7 and 8.

6.1.2

Results from PROTEUS Software

SPWM signal generation from microcontroller: The PIC microcontroller is taken for generation of SPWM signal. A program is written in assembly language in MPLAB IDE programming software. The SPWM signals of 5 V of peak magnitude is developed in PROTEUS Design suite software after Hex code of corresponding SPWM program is loaded into PIC microcontroller in PROTEUS software (Figs. 9 and 10).

Fig. 4 5-level voltage waveform for R-load

Design of a Multilevel Inverter Using SPWM Technique

Fig. 5 Current waveform for R-L load

Fig. 6 9-level voltage waveform for R-load

223

224

A. Ray et al.

Fig. 7 Current waveform for R-L load

6.2 Result Found from Hardware Circuit Design The hardware section has been designed for a testing purpose for the generation of 5-level inverter output voltage. The generated Hex file of SPWM program loaded into PIC microcontroller for generation of SPWM signal is responsible for triggering of inverter switches. The program has also been written in that fashion to change 5-Level output voltage to 3-Level output voltage. The DC power supply is given as Vdc = 5 V. The design was tested on a pure resistive load (Figs. 11 and 12). Analysis of total harmonic distortion (THD) of output voltage waveform has been performed in MATLAB software. The result is shown in Table 2. THD level is improved for higher level inverter compared to low-level inverter.

7 Conclusion and Future Scope of Work The multilevel inverter can be designed for different kind of applications. For the generation of different levels of output voltage only one PWM generator module of PIC microcontroller is sufficient. After the literature survey, we found for “m” level multilevel inverter the minimum number of carrier signal required is m − 1. But from our design, we can easily say only one carrier signal is sufficient for generation of

Design of a Multilevel Inverter Using SPWM Technique

225

Fig. 8 3-level output voltage waveform from 5-level inverter circuit

“m” level multilevel inverter. The condition is to use our microcontroller in a different manner. So we can easily save PWM module of a particular microcontroller for generation of multilevel output voltage waveform. As microcontroller has independent PWM modules, so we can save PWM module for serving some other purposes. As we have used only one PWM module using an independent timer so other timers are free for accessing other interrupts. If the program of SPWM generation is updated by changing some particular instructions then we can generate lower level output voltage by using the higher order inverter circuit, i.e., it is possible to generate 3level output voltage from 5-level circuit design. Application wise it may require to convert the design to lower level as the power supply is a major consideration for the design of multilevel inverter. Total harmonic distortion is a criterion for expressing the performance of an inverter. If we will increase the output voltage levels then the THD level decrease accordingly. So if we will be able to generate more number of output voltage levels then THD level will decrease accordingly. We usually prefer multilevel inverter for medium voltage high power application so the design can be modified in 3-ph system.

226

Fig. 9 SPWM signal for 5-level inverter in PROTEUS software

Fig. 10 SPWM Signal for 3-level inverter in PROTEUS

A. Ray et al.

Design of a Multilevel Inverter Using SPWM Technique

227

Fig. 11 5-level output voltage waveform from hardware design

We can bring DC capacitors in the input side of our designed circuit rather use separate DC batteries. The DC capacitors voltage unbalancing is the main technical drawback of a diode clamped multilevel inverter. A voltage balancing circuit based on buck–boost chopper connected to the DC link capacitor of diode clamped multilevel inverter is a reliable and robust solution of this problem. And overall circuit can be modified further for its future applications like: industry/traction/FACTS devices.

228

A. Ray et al.

Fig. 12 3-level output voltage waveform from hardware design

Table 2 Analysis of THD of output voltage waveform

Voltage level

THD (%)

3-level

36.16

5-level

31.68

9-level

23.51

References 1. Kouro S, Gopakumar K, Malinowski M (2010) Recent advances and industrial application of multilevel converter. IEEE Trans Ind Electron 57(8):2553–2580 2. De S, Banerjee D, Sivakumar K, Gopakumar K, Ramchand R, Patel C (2011) Multilevel inverters for low-power application. IET Power Electron 4(4):384–392 3. Akagi H, Fujita H, Yonetani S (2008) A 6.6-kV transformerless STATCOM based on a fivelevel diode-clamped PWM converter: system design and experimentation of a 200-V 10- kVA laboratory model. IEEE Trans Ind Appl 44(2):672–680 4. Cheng Y, Qian C, Crow ML, Pekarek S (2006) A comparison of diode-clamped and cascaded multilevel converters for a STATCOM with energy storage. IEEE Trans Ind Electron 53(5):1512–1521 5. Williamson SS, Woronowicz K (2016) Design and development of an efficient multilevel DC/AC traction inverter for railway transportation electrification. IEEE Trans Power Electron 31(4):3036–3042 6. Lai JS, Peng FZ (1996) Multilevel converters-a new breed of power converters. IEEE Trans Ind Appl 32(3):509–516 7. Bernet S (2000) Recent developments of high power converters for industry and traction applications. IEEE Trans Power Electron 15(6):1102–1117

Design of a Multilevel Inverter Using SPWM Technique

229

8. Farokhnia N, Fathi SH, Vadizadeh H, Toodeji H (2010) Comparison between approximate and accurate calculation of voltage THD in multilevel inverters with unequal DC sources. In: Proceedings of 5th IEEE industrial electronics and applications conference, June 2010, pp 1034–1039 9. Yuan X, Barbi I (2000) Fundamentals of a new diode clamping multilevel inverter. IEEE Trans Power Electron 15(4):711–718 10. Rodríguez J, Lai JS, Peng FZ (2002) Multilevel inverters: a survey of topologies, controls, and applications. IEEE Trans Ind Electron 49(4):724–738